O'Reilly Data Show Podcast: Recent Episodes
O'Reilly Media
The O'Reilly Data Show Podcast explores the opportunities and techniques driving big data, data science, and AI.
In this episode of the Data Show, I speak with Peter Bailis, founder and CEO of Sisu, a startup that is using machine learning to improve operational analytics. Bailis is also an assistant professor of computer science at Stanford University, where he conducts research into data-intensive systems and where he is co-founder of the DAWN […]
In this episode of the Data Show, I speak with Arun Kejariwal of Facebook and Ira Cohen of Anodot (full disclosure: I’m an advisor to Anodot). This conversation stemmed from a recent online panel discussion we did, where we discussed time series data, and, specifically, anomaly detection and forecasting. Both Kejariwal (at Machine Zone, Twitter, […]
In this episode of the Data Show, I speak with Michael Mahoney, a member of RISELab, the International Computer Science Institute, and the Department of Statistics at UC Berkeley. A physicist by training, Mahoney has been at the forefront of many important problems in large-scale data analysis. On the theoretical side, his works spans algorithmic […]
In this episode of the Data Show, I speak with Kesha Williams, technical instructor at A Cloud Guru, a training company focused on cloud computing. As a full stack web developer, Williams became intrigued by machine learning and started teaching herself the ML tools on Amazon Web Services. Fast forward to today, Williams has built […]
In this episode of the Data Show, I speak with Alex Ratner, project lead for Stanford’s Snorkel open source project; Ratner also recently garnered a faculty position at the University of Washington and is currently working on a company supporting and extending the Snorkel project. Snorkel is a framework for building and managing training data. […]
In this episode of the Data Show, I speak with Cassie Kozyrkov, technical director and chief decision scientist at Google Cloud. She describes “decision intelligence” as an interdisciplinary field concerned with all aspects of decision-making, and which combines data science with the behavioral sciences. Most recently she has been focused on developing best practices that […]
In this episode of the Data Show, I spoke with Roger Chen, co-founder and CEO of Computable Labs, a startup focused on building tools for the creation of data networks and data exchanges. Chen has also served as co-chair of O’Reilly’s Artificial Intelligence Conference since its inception in 2016. This conversation took place the day […]
In this week’s episode of the Data Show, we’re featuring an interview Data Show host Ben Lorica participated in for the Software Engineering Daily Podcast, where he was interviewed by Jeff Meyerson. Their conversation mainly centered around data engineering, data architecture and infrastructure, and machine learning (ML). Here are a few highlights: Tools for productive […]
In this episode of the Data Show, I spoke with Nick Pentreath, principal engineer at IBM. Pentreath was an early and avid user of Apache Spark, and he subsequently became a Spark committer and PMC member. Most recently his focus has been on machine learning, particularly deep learning, and he is part of a group […]
In this episode of the Data Show, I spoke with Dhruba Borthakur (co-founder and CTO) and Shruti Bhat (SVP of Product) of Rockset, a startup focused on building solutions for interactive data science and live applications. Borthakur was the founding engineer of HDFS and creator of RocksDB, while Bhat is an experienced product and marketing […]
In this episode of the Data Show, I spoke with Jike Chong, chief data scientist at Acorns, a startup focused on building tools for micro-investing. Chong has extensive experience using analytics and machine learning in financial services, and he has experience building data science teams in the U.S. and in China. We had a great […]
In this episode of the Data Show, I spoke with Jeff Jonas, CEO, founder and chief scientist of Senzing, a startup focused on making real-time entity resolution technologies broadly accessible. He was previously a fellow and chief scientist of context computing at IBM. Entity resolution (ER) refers to techniques and tools for identifying and linking […]
In this episode of the Data Show, I spoke with Neelesh Salian, software engineer at Stitch Fix, a company that combines machine learning and human expertise to personalize shopping. As companies integrate machine learning into their products and systems, there are important foundational technologies that come into play. This shouldn’t come as a shock, as […]
In this episode of the Data Show, I spoke with Avner Braverman, co-founder and CEO of Binaris, a startup that aims to bring serverless to web-scale and enterprise applications. This conversation took place shortly after the release of a seminal paper from UC Berkeley (“Cloud Programming Simplified: A Berkeley View on Serverless Computing”), and this […]
In this episode of the Data Show, I spoke with Forough Poursabzi-Sangdeh, a postdoctoral researcher at Microsoft Research New York City. Poursabzi works in the interdisciplinary area of interpretable and interactive machine learning. As models and algorithms become more widespread, many important considerations are becoming active research areas: fairness and bias, safety and reliability, security […]
In this episode of the Data Show, I spoke with Kartik Hosanagar, professor of technology and digital business, and professor of marketing at The Wharton School of the University of Pennsylvania. Hosanagar is also the author of a newly released book, A Human’s Guide to Machine Intelligence, an interesting tour through the recent evolution of AI applications that draws from his extensive experience at the intersection of business and technology.
We had a great conversation spanning many topics, including:
- The types of unanticipated consequences of which algorithm designers should be aware.
- The predictability-resilience paradox: as systems become more intelligent and dynamic, they also become more unpredictable, so there are trade-offs algorithms designers must face.
- Managing risk in machine learning: AI application designers need to weigh considerations such as fairness, security, privacy, explainability, safety, and reliability.
- A bill of rights for humans impacted by the growing power and sophistication of algorithms.
- Some best practices for bringing AI into the enterprise.
Related resources:
- “Managing risk in machine learning”: considerations for a world where ML models are becoming mission critical
- Francesca Lazzeri and Jaya Mathew on “Lessons learned while helping enterprises adopt machine learning”
- Jerry Overton on “Teaching and implementing data science and AI in the enterprise”
- Kris Hammond on “Bringing AI into the enterprise”
- Jacob Ward on “How social science research can inform the design of AI systems”
- “Overcoming barriers to AI adoption”
- Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models”
In this episode of the Data Show, I spoke with P.W. Singer, strategist and senior fellow at the New America Foundation, and a contributing editor at Popular Science. He is co-author of an excellent new book, LikeWar: The Weaponization of Social Media, which explores how social media has changed war, politics, and business. The book is essential reading for anyone interested in how social media has become an important new battlefield in a diverse set of domains and settings.
We had a great conversation spanning many topics, including:
- In light of the 10th anniversary of his earlier book Wired for War, we talked about progress in robotics over the past decade.
- The challenge posed by the fact that social networks reward virality, not veracity.
- How the internet has emerged as an important new battlefield.
- How this new online battlefield changes how conflicts are fought and unfold.
- How many of the ideas and techniques covered in LikeWar are trickling down from nation-state actors influencing global events, to consulting companies offering services that companies and individuals can use.
Here are some highlights from our conversation:
LikeWar
We spent five years tracking how social media was being used all around the world. … We looked at everything from how was it being used by militaries, by terrorist groups, by politicians, by teenagers—you name it. The finding of this project is sort of a two-fold play on words. The first is, if you think of cyberwar as the hacking of networks, LikeWar is its twin. It’s the hacking of people on the networks by driving ideas viral through a mix of likes and lies.
… Social media began as a space for fun, for entertainment. It then became a communication space. It became a marketplace. It’s also turned it into a kind of battle space. It’s simultaneously all of these things at once, and you can see, for example, Russian information warriors who are using digital marketing techniques and teenage jokes to influence the outcomes of elections. A different example would be ISIS’ top recruiter, Junaid Hussain, mimicking how Taylor Swift built her fan army.
A common set of tactics
The second finding of the project was that when you look across all these wildly diverse actors, groups, and organizations, they turned out to be using very similar tactics, very similar approaches. To put it a different way: it’s a mode of conflict. There’s ways of “winning” that all the different groups are realizing. More importantly, the groups that understand these new rules of the game are the ones that are winning their online wars and having a real effect, whether that real effect is winning a political campaign, winning a corporate marketing campaign, winning a campaign to become a celebrity, or to become the most popular kid in school. Or “winning” might be to do the opposite—to sabotage someone else’s campaign to become a leading political candidate.
Related resources:
- Siwei Lyu on “The technical, societal, and cultural challenges that come with the rise of fake media”
- Supasorn Suwajanakorn on “Building artificial people: Endless possibilities and the dark side”
- Guillaume Chaslot on “The importance of transparency and user control in machine learning”
- “Overcoming barriers to AI adoption”
- Alon Kaufman on “Machine learning on encrypted data”
- Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models”
In this episode of the Data Show, I spoke with Siwei Lyu, associate professor of computer science at the University at Albany, State University of New York. Lyu is a leading expert in digital media forensics, a field of research into tools and techniques for analyzing the authenticity of media files. Over the past year, there have been many stories written about the rise of tools for creating fake media (mainly images, video, audio files). Researchers in digital image forensics haven’t exactly been standing still, though. As Lyu notes, advances in machine learning and deep learning have also found a receptive audience among the forensics community.
We had a great conversation spanning many topics including:
- The many indicators used by forensic experts and forgery detection systems
- Balancing “open” research with risks that come with it—including “tipping off” adversaries
- State-of-the-art detection tools today, and what the research community and funding agencies are working on over the next few years.
- Technical, societal, and cultural challenges that come with the rise of fake media.
Here are some highlights from our conversation:
Imbalance between digital forensics researchers and forgers
In theory, it looks difficult to synthesize media. This is true, but on the other hand, there are factors to consider on the side of the forgers. The first is the fact that most people working in forensics, like myself, usually just write a paper and publish it. So, the details of our detection algorithm becomes available immediately. On the other hand, people making fake media are usually secretive; they don’t usually publish the details of their algorithms. So, there’s a kind of imbalance between the information on the forensic side and the forgery side.
The other issue is user habit. The fact that even if some of the fakes are very low quality, a typical user checks it just for a second; sees something interesting, exciting, sensational; and helps distribute it without actually checking the authenticity. This actually helps fake media to broadcast very, very fast. Even though we have algorithms to detect fake media, these tools are probably not fast enough to actually stop the trap.
… Then there are the actual incentives for this kind of work. For forensics, even if we have the tools and the time to catch a piece of fake media, we don’t get anything. But for people actually making the fake media, there is more financial or other forms of incentive to do that.
Related resources:
- Supasorn Suwajanakorn on “Building artificial people: Endless possibilities and the dark side”
- Alyosha Efros on “Using computer vision to understand big visual data”
- “Overcoming barriers to AI adoption”
- “What is neural architecture search?”
- Alon Kaufman on “Machine learning on encrypted data”
- Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models”
In this episode of the Data Show, I spoke with Maryam Jahanshahi, research scientist at TapRecruit, a startup that uses machine learning and analytics to help companies recruit more effectively. In an upcoming survey, we found that a “skills gap” or “lack of skilled people” was one of the main bottlenecks holding back adoption of AI technologies. Many companies are exploring a variety of internal and external programs to train staff on new tools and processes. The other route is to hire new talent. But recent reports suggest that demand for data professionals is strong and competition for experienced talent is fierce. Jahanshahi and her team are building natural language and statistical tools that can help companies improve their ability to attract and retain talent across many key areas.
Here are some highlights from our conversation:
Optimal job titles
The conventional wisdom in our field has always been that you want to optimize for “the number of good candidates” divided by “the number of total candidates.” … The thinking is that one of the ways in which you get a good signal-to-noise ratio is if you advertise for a more senior role. … In fact, we found the number of qualified applicants was lower for the senior data scientist role.
… We saw from some of our behavioral experiments that people were feeling like that was too senior a role for them to apply to. What we would call the “confidence gap” was kicking in at that point. It’s a pretty well-known phenomena that there are different groups of the population that are less confident. This has been best characterized in terms of gender. It’s the idea that most women only apply for jobs when they meet 100% of the qualifications versus most men will apply even with 60% of the qualifications. That was actually manifesting.
Highlighting benefits
We saw a lot of big companies that would offer 401(k), that would offer health insurance or family leave, but wouldn’t mention those benefits in the job descriptions. This had an impact on how candidates perceived these companies. Even though it’s implied that Coca-Cola is probably going to give you 401(k) and health insurance, not mentioning it changes the way you think of that job.
… So, don’t forget the things that really should be there. Even the boring stuff really matters for most candidates. You’d think it would only matter for older candidates, but, actually, millennials and everyone in every age group are very concerned about these things because it’s not specifically about the 401(k) plan; it’s about what it implies in terms of the company—that the company is going to take care of you, is going to give you leave, is going to provide a good workplace.
Improving diversity
We found the best way to deal with representation at the end of the process is actually to deal with representation early in the process. What I mean by that is having a robust or a healthy candidate pool at the start of the process. We found for data scientist roles, that was about having 100 candidates apply for your job.
… If we’re not getting to the point where we can attract 100 applicants, we’ll take a look at that job description. We’ll see what’s wrong with it and what could be turning off candidates; it could be that you’re not syndicating the job description well, it’s not getting into search results, or it could be that it’s actually turning off a lot of people. You could be asking for too many qualifications, and that turns off a lot of people. … Sometimes it involves taking a step back and taking a look at what we’re doing in this process that’s not helping us and that’s starving us of candidates.
Related resources:
- Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models”
- “Comparing production-grade NLP libraries”
- “What are machine learning engineers?”
- “The next generation of AI assistants in enterprise”
- David Blei on “Topic models: Past, present, and future”
- David Ferrucci on why “Language understanding remains one of AI’s grand challenges”
In this episode of the Data Show, I spoke with Andrew Burt, chief privacy officer and legal engineer at Immuta, a company building data management tools tuned for data science. Burt and cybersecurity pioneer Daniel Geer recently released a must-read white paper (“Flat Light”) that provides a great framework for how to think about information security in the age of big data and AI. They list important changes to the information landscape and offer suggestions on how to alleviate some of the new risks introduced by the rise of machine learning and AI.
We discussed their new white paper, cybersecurity (Burt was previously a special advisor at the FBI), and an exciting new Strata Data tutorial that Burt will be co-teaching in March.
Privacy and security are converging
The end goal of privacy and the end goal of security are now totally summed up by this idea: how do we control data in an environment where that control is harder and harder to achieve, and in an environment that is harder and harder to understand?
… As we see machine learning become more prominent, what’s going to be really fascinating is that, traditionally, both privacy and security are really related to different types of access. One was adversarial access in the case of security; the other is the party you’re giving the data to accessing it in a way that aligns with your expectations—that would be a traditional notion of privacy. … What we’re going to start to see is that both fields are going to be more and more worried about unintended entrances.
Data lineage and data provenance
One of the things we say in the paper is that as we move to a world where models and machine learning increasingly take the place of logical instruction-oriented programming, we’re going to have less and less source code, and we’re going to have more and more source data. And as that shift occurs, what then becomes most important is understanding everything we can about where that data came from, who touched it, and if its integrity has in fact been preserved.
In the white paper, we talk about how, when we think about integrity in this world of machine learning and models, it does us a disservice to think about a binary state, which is the traditional way: “either data is correct or it isn’t. Either it’s been tampered with, or it hasn’t been tampered with.” And that was really the measure by which we judged whether failures had occurred. But when we’re thinking not about source code but about source data for models, we need to be moving into more of a probabilistic mode. Because when we’re thinking about data, data in itself is never going to be fully accurate. It’s only going to be representative to some degree of whatever it’s actually trying to represent.
Related resources:
- “Managing risk in machine learning”
- Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models”
- Alon Kaufman on “Machine learning on encrypted data”
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain.
- “We need to build machine learning tools to augment machine learning engineers”
- “Case studies in data ethics”
In this episode of the Data Show, I spoke with Haoyuan Li, CEO and founder of Alluxio, a startup commercializing the open source project with the same name (full disclosure: I’m an advisor to Alluxio). Our discussion focuses on the state of Alluxio (the open source project that has roots in UC Berkeley’s AMPLab), specifically emerging use cases here and in China. Given the large-scale use in China, I also wanted to get Li’s take on the state of data and AI technologies in Beijing and other parts of China.
Here are some highlights from our conversation:
A much needed layer between compute and storage in a world with disparate storage systems
This new layer, which we call a virtual distributed file system, sits in the middle between the compute and storage layers. This new layer virtualizes data from different storage systems and presents a unified API with a global namespace for the data-driven applications to interact with all of the data in the enterprise environment.
AI and machine learning applications
One key reason people use an object store is that it is cheap. Per gigabyte or per terabyte, it’s cheaper than other solutions in a market,…but performance is not as good. And from that perspective, by putting open source Alluxio on top of that, that improves performance from Alluxio’s caching functionality. On top of that, in many cases, machine learning libraries cannot directly talk with object stores, and Alluxio can also serve as a translation layer.
Adoption in China
Things are moving very fast in that region. People are eager to adopt new technology, particularly for AI and big data. Some are users we know very quickly boosted their Alluxio deployments to hundreds of nodes or even thousands of nodes. It’s amazing to see how fast they can adapt.
… Of the top 10 internet companies in China, nine are using open source Alluxio in production today. All nine of them have big data and AI use cases for Alluxio. … I also travel back and forth between these two regions quite often, and every time I go there, I see more use cases, more applications, and more innovation.
Related resources:
- Michael Franklin on the lasting legacy of AMPLab
- Jason Dai on why “Companies in China are moving quickly to embrace AI technologies”
- Kai-Fu Lee on “China: AI superpower”
- Andrew Feldman on why “Specialized hardware for deep learning will unleash innovation”
- Greg Diamos on “How big compute is powering the deep learning rocket ship”
- Tim Kraska on “How machine learning will accelerate data management systems”
For the end-of-year holiday episode of the Data Show, I turned the tables on Data Show host Ben Lorica to talk about trends in big data, machine learning, and AI, and what to look for in 2019. Lorica also showcased some highlights from our upcoming Strata Data and Artificial Intelligence conferences.
Here are some highlights from our conversation:
Real-world use cases for new technology
If you’re someone who wants to use data, data infrastructure, data science, machine learning, and AI, we’re really at the point where there are a lot of tools for implementers and developers. They’re not necessarily doing research and development; they just want to build better products and automate workflow. I think that’s the most significant development in my mind.
And then I think use case sharing also has an impact. For example, at our conferences, people are sharing how they’re using AI and ML in their businesses, so the use cases are getting better defined—particularly for some of these technologies that are relatively new to the broader data community, like deep learning. There are now use cases that touch the types of problems people normally tackle—so, things that involve structured data, for example, for time series forecasting, or recommenders.
With that said, while we are in an implementation phase, I think as people who follow this space will attest, there’s still a lot of interesting things coming out of the R&D world, so still a lot of great innovation and a lot more growth in terms of how sophisticated and how easy to use these technologies will be.
Addressing ML and AI bottlenecks
We have a couple of surveys that we’ll release early in 2019. In one of these surveys, we asked people what the main bottleneck is in terms of adopting machine learning and AI technologies.
Interestingly enough, the main bottleneck was cultural issues—people are still facing challenges in terms of convincing people within their companies to adopt these technologies. And then, of course, the next two are the ones we’re familiar with: lack of data and lack of skilled people. And then the fourth bottleneck people cited was trouble identifying business use cases.
What’s interesting about that is, if you then ask people how mature their practice is and you look at the people with the most mature AI and machine learning practices, they still cite a lack of data as the main bottleneck. What that tells me is that there’s still a lot of opportunity for people to apply these technologies within their companies, but there’s a lot of foundational work people have to do in terms of just getting data in place, getting data collected and ready for analytics.
Focus on foundational technologies
At the Strata Data conferences in San Francisco, London, and New York, the emphasis will be building technologies, bringing in technologies and cultural practices that will allow you to sustain analytics and machine learning in your organization. That means having all of the foundational technologies in place—data ingestion, data governance, ETL, data lineage, data science platform, metadata, store, and things like that, the various pieces of technology that will be important as you scale the practice of machine learning and AI in your company.
At the Artificial Intelligence conferences, we remain focused on being the de facto gathering place for people interested in applied artificial intelligence. We will focus on servicing the most important use cases in many, many domains. That means showcasing, of course, the latest research in deep learning and other branches of machine learning, but also helping people grapple with some of the other important considerations, like privacy and security, fairness, reliability, and safety.
…At both the Strata Data and Artificial Intelligence conferences, we will focus on helping people understand the capabilities of the technology, the strengths and limitations; that’s why we run executive briefings at all of these events. We showcase case studies that are aimed at the non-technical and business user as well—so, we’ll have two types of case studies, one more technical and one not so technical so the business decision-makers can benefit from seeing how their peers are using and succeeding with some of these technologies.
In this episode of the Data Show, I spoke with Alex Wong, associate professor at the University of Waterloo, and co-founder of DarwinAI, a startup that uses AI to address foundational challenges with deep learning in the enterprise. As the use of machine learning and analytics become more widespread, we’re beginning to see tools that enable data scientists and data engineers to scale and tackle many more problems and maintain more systems. This includes automation tools for the many stages involved in data science, including data preparation, feature engineering, model selection, and hyperparameter tuning, as well as tools for data engineering and data operations.
Wong and his collaborators are building solutions for enterprises, including tools for generating efficient neural networks and for the performance analysis of networks deployed to edge devices.
Here are some highlights from our conversation:
Using AI to democratize deep learning
Having worked in machine learning and deep learning for more than a decade, both in academia as well as industry, it really became very evident to me that there’s a significant barrier to widespread adoption. One of the main things is that it is very difficult to design, build, and explain deep neural networks. I especially wanted to meet operational requirements. The process just involves way too much guesswork, trial and error, so it’s hard to build systems that work in real-world industrial systems.
One of the out-of-the-box moments we had—pretty much the only way we could actually do this—was to reinvent the way we think about building deep neural networks. Which is, can we actually leverage AI itself as a collaborative technology? Can we build something that works with people to design and build much better networks? And that led to the start of DarwinAI—our main vision is pretty much enabling deep learning for anyone, anywhere, anytime.
Generative synthesis
The general concept of generative synthesis is to find the best generative model that meets your particular operational requirements (which could be size, speed, accuracy, and so forth). So, the intuition behind that is that we treat it as a large constrained optimization problem where we try to identify the generative machine that will actually give you the highest performance. We have a unique way of having an interplay between a generator and an inquisitor where the generator will generate networks that the inquisitor probes and understands. Then it learns intuition about what makes a good network and what doesn’t.
Related resources:
- Vitaly Gordon on “Building tools for enterprise data science”
- “What machine learning means for software development”
- “We need to build machine learning tools to augment machine learning engineers”
- Tim Kraska on “How machine learning will accelerate data management systems”
- “Building tools for the AI applications of tomorrow”
In this episode of the Data Show, I spoke with Vitaly Gordon, VP of data science and engineering at Salesforce. As the use of machine learning becomes more widespread, we need tools that will allow data scientists to scale so they can tackle many more problems and help many more people. We need automation tools for the many stages involved in data science, including data preparation, feature engineering, model selection and hyperparameter tuning, as well as monitoring.
I wanted the perspective of someone who is already faced with having to support many models in production. The proliferation of models is still a theoretical consideration for many data science teams, but Gordon and his colleagues at Salesforce already support hundreds of thousands of customers who need custom models built on custom data. They recently took their learnings public and open sourced TransmogrifAI, a library for automated machine learning for structured data, which sits on top of Apache Spark.
Here are some highlights from our conversation:
The need for an internal data science platform
It’s more about how much commonality there is between every single data science use case—how many of the problems are redundant and repeatable.
… A lot of data scientists solve problems that honestly have a lot to do with engineering, a lot to do with things that are not pure modeling.
TransmogrifAI
TransmogrifAI is an automated machine library for mostly structured data, and the problem that it aims to solve is that we at Salesforce have hundreds of thousands of customers. While all of them share a common set of data, the Salesforce platform itself is extremely customizable. Actually, 80% of the data inside the Salesforce platform actually sits in what we refer to as custom objects, which one can think of as custom tables in a database.
… We don’t build models that are shared between customers. We always use a single customer’s data. We have hundreds of thousands of models potentially that we need to build, and because of that, we needed to automate the entire process. We just cannot throw people at the problem. We basically created TransmogrifAI to automate the entire end-to-end process for creating a model for a user and we decided to open source it a couple months ago.
Related resources:
- “What machine learning means for software development”
- “We need to build machine learning tools to augment machine learning engineers”
- Francesca Lazzeri and Jaya Mathew on “Lessons learned while helping enterprises adopt machine learning”
- Tim Kraska on “How machine learning will accelerate data management systems”
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain.
- “Lessons learned turning machine learning models into real products and services”
In this episode of the Data Show, I spoke with Francesca Lazzeri, an AI and machine learning scientist at Microsoft, and her colleague Jaya Mathew, a senior data scientist at Microsoft. We conducted a couple of surveys this year—“How Companies Are Putting AI to Work Through Deep Learning” and “The State of Machine Learning Adoption in the Enterprise”—and we found that while many companies are still in the early stages of machine learning adoption, there’s considerable interest in moving forward with projects in the near future. Lazzeri and Mathew spend a considerable amount of time interacting with companies that are beginning to use machine learning and have experiences that span many different industries and applications. I wanted to learn some of the processes and tools they use when they assist companies in beginning their machine learning journeys.
Here are some highlights from our conversation:
Team data science process
Francesca Lazzeri: The Data Science Process is a framework that we try to apply in our projects. Everything begins with a business problem, so external customers come to us with a business problem or a process they want to optimize. We work with them to translate these into realistic questions, into what we call data science questions. And then we move to the data portion: what are the different relevant data sources, is the data internal or external? After that, you try to define the data pipeline. We start with the core part of the data science process—that is, data cleaning—and proceed to feature engineering, model building, and model deployment and management.
…There are also usually external agents involved. When I say external agents, I mean there are program managers and business experts who follow us during this process. These are individuals who are the data and domain experts. It’s a very interactive process because you go back and forth trying to understand if what you are building is something that really can be interesting to the business owners.
What is holding back adoption of machine learning
Jaya Mathew: One of the biggest bottlenecks is lack of talent within the organization. A company really needs to invest in either up-scaling their existing employee base, which tends to be expensive and they’re trying to figure out if that investment is really worth it. Or they need to try to hire, and hiring specific skill sets is difficult, as there is a talent shortage everywhere.
Then, in addition to that, there’s also a little bit of hesitation because some of the AI and machine learning models are “black boxes”. … I think many governments and many organizations need to be able to explain what’s going on before they deploy a model.
Related resources:
- Francesca Lazzeri and Jaya Mathew: “A day in the life of a data scientist: How do we train our teams to get started with AI?”
- Ashok Srivastava on why “The real value of data requires a holistic view of the end-to-end data pipeline”
- Jerry Overton on “Teaching and implementing data science and AI in the enterprise”
- Carme Artigas on “Transforming organizations through analytics centers of excellence”
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain.
In this episode of the Data Show, I spoke with Alon Kaufman, CEO and co-founder of Duality Technologies, a startup building tools that will allow companies to apply analytics and machine learning to encrypted data. In a recent talk, I described the importance of data, various methods for estimating the value of data, and emerging tools for incentivizing data sharing across organizations. As I noted, the main motivation for improving data liquidity is the growing importance of machine learning. We’re all familiar with the importance of data security and privacy, but probably not as many people are aware of the emerging set of tools at the intersection of machine learning and security. Kaufman and his stellar roster of co-founders are doing some of the most interesting work in this area.
Here are some highlights from our conversation:
Running machine learning models on encrypted data
Four or five years ago, techniques for running machine learning models on data while it’s encrypted were being discussed in the academic world. We did a few trials of this and although the results were fascinating, it still wasn’t practical.
… There have been big breakthroughs that have led to it becoming feasible. A few years ago, it was more theoretical. Now it’s becoming feasible. This is the right time to build a company. Not only because of the technology feasibility but definitely because of the need in the market.
From inference to training
A classical example would be model inference. I have data; you have some predictive model. I want to consume your model. I’m not willing to share my data with you, so I’ll encrypt my data; you’ll apply your model to the encrypted data, so you’ll never see the data. I will never see your model. The result that comes out of this computation, which is encrypted as well, will be decrypted only by me, as I have the key. This means I can basically utilize your predictive insight, you can sell your model, and no one ever exchanged data or models between the parties.
… The next frontier of research is doing model training with these type of technologies. We have some great results, and there are others who are starting to do and implement some things in hardware. … Some of our recent work around applying deep learning to encrypted data combines different methods. Homomorphic encryption has its pros and cons; secure multi-party computation has other advantages and disadvantages. We basically mash various methods together to derive very, very interesting results. … For example, we have applied algorithms to genomic data at scale and we obtained impressive performance.
Related resources:
- Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models”
- Chang Liu on “How privacy-preserving techniques can lead to more robust machine learning models”
- “How to build analytic products in an age when data privacy has become critical”
- “Data collection and data markets in the age of privacy and machine learning”
- “What machine learning means for software development”
- “Lessons learned turning machine learning models into real products and services”
In this episode of the Data Show, I spoke with Jacob Ward, a Berggruen Fellow at Stanford University. Ward has an extensive background in journalism, mainly covering topics in science and technology, at National Geographic, Al Jazeera, Discovery Channel, BBC, Popular Science, and many other outlets. Most recently, he’s become interested in the interplay between research in psychology, decision-making, and AI systems. He’s in the process of writing a book on these topics, and was gracious enough to give an informal preview by way of this podcast conversation.
Here are some highlights from our conversation:
Psychology and AI
I began to realize there was a disconnect between what is a totally revolutionary set of innovations coming through in psychology right now that are really just beginning to scratch the surface of how human beings make decisions; at the same time, we are beginning to automate human decision-making in a really fundamental way. I had a number of different people say, ‘Wow, what you’re describing in psychology really reminds me of this piece of AI that I’m building right now,’ to change how expectant mothers see their doctors or change how we hire somebody for a job or whatever it is.
Transparency and designing systems that are fair
I was talking to somebody the other day who was trying to build a loan company that was using machine learning to present loans to people. He and his company did everything they possibly could to not redline the people they were loaning to. They were trying very hard not to make unfair loans that would give preference to white people over people of color.
They went to extraordinary lengths to make that happen. They cut addresses out of the process. They did all of this stuff to try to basically neutralize the process, and the machine learning model still would pick white people at a disproportionate rate over everybody else. They can’t explain why. They don’t know why that is. There’s some variable that’s mapping to race that they just don’t know about.
But that sort of opacity—this is somebody explaining it to me who just happened to have been inside the company, but it’s not as if that’s on display for everybody to check out. These kinds of closed systems are picking up patterns we can’t explain, and that their creators can’t explain. They are also making really, really important decisions based on them. I think it is going to be very important to change how we inspect these systems before we begin trusting them.
Anthropomorphism and complex systems
In this book, I’m also trying to look at the way human beings respond to being given an answer by an automated system. There are some very well-established, psychological principles out there that can give us some sense of how people are going to respond when they are told what to do based on an algorithm.
The people who study anthropomorphism, the imparting of intention and human attributes to an automated system, say there’s a really well-established pattern. When people are shown a very complex system and given some sort of exposure to that complex system, whether it gives them an answer or whatever it is, it tends to produce in human beings a level of trust in that system that doesn’t really have anything to do with reality. … The more complex the system, the more people tend to trust it.
Related resources:
- Jacob Ward on “How AI will amplify the best and worst of humanity”
- Sharad Goel and Sam Corbett-Davies on “Why it’s hard to design fair machine learning models”
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain.
- “We need to build machine learning tools to augment machine learning engineers”
- “Case studies in data ethics”
- “Haunted by data”: Maciej Ceglowski makes the case for adopting enforceable limits for data storage.
In this episode of the Data Show, I spoke with Sharad Goel, assistant professor at Stanford, and his student Sam Corbett-Davies. They recently wrote a survey paper, “A Critical Review of Fair Machine Learning,” where they carefully examined the standard statistical tools used to check for fairness in machine learning models. It turns out that each of the standard approaches (anti-classification, classification parity, and calibration) has limitations, and their paper is a must-read tour through recent research in designing fair algorithms. We talked about their key findings, and, most importantly, I pressed them to list a few best practices that analysts and industrial data scientists might want to consider.
Here are some highlights from our conversation:
Calibration and other standard metrics
Sam Corbett-Davies: The problem with many of the standard metrics is that they fail to take into account how different groups might have different distributions of risk. In particular, if there are people who are very low risk or very high risk, then it can throw off these measures in a way that doesn’t actually change what the fair decision should be. … The upshot is that if you end up enforcing or trying to enforce one of these measures, if you try to equalize false positive rates, or you try to equalize some other classification parity metric, you can end up hurting both the group you’re trying to protect and any other groups for which you might be changing the policy.
… A layman’s definition of calibration would be, if an algorithm gives a risk score—maybe it gives a score from one to 10, and one is very low risk and 10 is very high risk—calibration says the scores should mean the same thing for different groups (where the groups are defined based on some protected variable like gender, age, or race). We basically say in our paper that calibration is necessary for fairness, but it’s not good enough. Just because your scores are calibrated doesn’t mean you aren’t doing something funny that could be harming certain groups.
The need to interrogate data
Sharad Goel: One way to operationalize this is if you have a set of reasonable measures to be your label, you can see how much your algorithm changes if you use different measures. If your algorithm is changing a lot using these different measures, then you really have to worry about determining the right measure. What is the right thing to predict. If it’s the case that under a variety of reasonable measures everything looks kind of stable, maybe it’s less of an issue. This is very hard to carry out in practice, but I do think it’s one of the most important things to understand and to be aware of when designing these types of algorithms.
… There are a lot of subtleties to these different types of metrics that are important to be aware of when designing these algorithms in an equitable way. … But fundamentally, these are hard problems. It’s not particularly surprising that we don’t have an algorithm to help us make all of these algorithms fair. … What is most important is that we really interrogate the data.
Related resources:
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain.
- “We need to build machine learning tools to augment machine learning engineers”
- “Case studies in data ethics”
- “Haunted by data”: Maciej Ceglowski makes the case for adopting enforceable limits for data storage.
In this episode of the Data Show, I spoke with Alan Nichol, co-founder and CTO of Rasa, a startup that builds open source tools to help developers and product teams build conversational applications. About 18 months ago, there was tremendous excitement and hype surrounding chatbots, and while things have quieted lately, companies and developers continue to refine and define tools for building conversational applications. We spoke about the current state of chatbots, specifically about the types of applications developers are building today and how he sees conversational applications evolving in the near future.
As I described in a recent post, workflow automation will happen in stages. With that in mind, chatbots and intelligent assistants are bound to improve as underlying algorithms, technologies, and training data get better.
Here are some highlights from our conversation:
Chatbots and state machines
The first component is what we call natural language understanding, which typically means taking a short message that a user sends and extracting some meaning from it, which means turning it into structured data. In the case we talked about regarding the SQL database, if somebody asks, for example, ‘What was my ROI on my Facebook campaigns last month?’, the first thing you want to understand is that this is a data question and you want to assign it a label identifying it as a person, and they’re not saying hello, or goodbye, or thank you, but asking a specific question. Then you want to pick out those fields to help you create a query.
… The second piece is, how do you actually know what to do next? How do you build a system that can hold a conversation that is coherent? What you realize very quickly is that it’s not enough to have one input always matched to the same output. For example, if you ask somebody a yes or no question and they say, ‘yes,’ the next thing to do, of course, depends on what the original question was.
… Real conversations aren’t stateless; they have some context and they need to pay attention to the history. So, the way developers do that is build a state machine. Which means, for example, that you have a bot that can do some different things. It can talk about flights; it can talk about hotels. Then you define different states for when the person is still searching, or for when they are comparing different things, or for when they finish a booking. And then you have to define rules for how to behave for every input, for every possible state.
Beyond state machines
The problem is that [the state machine] approach works for building your first version, but it really restricts you to what we call “the happy parts,” which is where the user is compliant and cooperative and does everything you ask them to do. But in typical cases, you ask a person, “Do you like option A, or option B?” Then you probably build the path for the person saying, A, you build a path for the person saying B. But then you give it to real users, and they say, “No, I don’t like either of those.” Or they ask a question like, “Why is A so much more expensive than B?” Or, “Let me get back to you about that.”
… They don’t scale, that’s the problem. If you’re a developer and somebody has a conversation with your bot and you realize that it did the wrong thing, now you have to go look back at your (literally) thousands or tens of thousands of rules to figure out which one crashed and which one did the wrong thing. You figure out where to inject one more rule to handle one more etiquette, and that just doesn’t scale at all.
… With our dialogue library Rasa core, we give the user the ability to talk to the bot and provide feedback. So, in Rasa, the whole flow of dialogue is also controlled with machine learning. And it’s learned from real sample conversations. You talk to the system and if it does something wrong, you provide feedback and it corrects itself. So, you explore the space of possible conversations interactively yourself, and then your users do as well.
Related resources:
- Alan Nichol on “The Next Generation of AI Assistants in Enterprise”
- “Bots: What you need to know”
- “Using machine learning to monitor and optimize chatbots”
- “Commercial speech recognition systems in the age of big data and deep learning”: Yishay Carmiel on applications of deep learning in text and speech
- “How to think about AI and machine learning technologies, and their roles in automation”
- “Deep learning revolutionizes conversational AI”
- “Topic models—past, present, and future”: David Blei discusses the origins and applications of topic models.
In this episode of the Data Show, I spoke with Eric Jonas, a postdoc in the new Berkeley Center for Computational Imaging. Jonas is also affiliated with UC Berkeley’s RISE Lab. It was at a RISE Lab event that he first announced Pywren, a framework that lets data enthusiasts proficient with Python run existing code at massive scale on Amazon Web Services. Jonas and his collaborators are working on a related project, NumPyWren, a system for linear algebra built on a serverless architecture. Their hope is that by lowering the barrier to large-scale (scientific) computation, we will see many more experiments and research projects from communities that have been unable to easily marshal massive compute resources. We talked about Bayesian machine learning, scientific computation, reinforcement learning, and his stint as an entrepreneur in the enterprise software space.
Here are some highlights from our conversation:
Pywren
The real enabling technology for us was when Amazon announced the availability of AWS Lambda, their microservices framework, in 2014. Following this prompting, I went home one weekend and thought, ‘I wonder how hard it is to take an arbitrary Python function and marshal it across the wire, get it running in Lambda; I wonder how many I can get at once?’ Thus, Pywren was born.
… Right now, we’re primarily focused on the entire scientific Python stack, so SciPy, NumPy, Pandas, Matplotlib, the whole ecosystem there. … One of the challenges with all of these frameworks and running these things on Lambda is that, right now, Lambda is a fairly constrained resource environment. Amazon will quite happily give you 3,000 cores in the next two seconds, but each one has a maximum runtime and a small amount of memory and a small amount of local disk. Part of the current active research thrust for Pywren is figuring out how to do more general-purpose computation within those resource limits. But right now, we mostly support everything you would encounter in your normal Python workflow—including Jupyter, NumPy, and scikit-learn.
Numpywren
Chris Ré has this nice quote: ‘Why is it easier to train a bidirectional LSTM with attention than it is to just compute the SVD of a giant matrix?’ One of these things is actually fantastically more complicated than the other, but right now, our linear algebra tools are just such an impediment to doing that sort of large-scale computation. We hope NumPyWren will enable this class of work for the machine learning community.
The growing importance of reinforcement learning
Ben Recht makes the argument that the most interesting problems in machine learning right now involve taking action based upon your intelligence. I think he’s right about this—taking action based upon past data and doing it in a way that is safe and robust and reliable and all of these sorts of things. That is very much the domain that has traditionally been occupied by fields like control theory and reinforcement learning.
Reinforcement learning and Ray
Ray is an excellent platform for building large-scale distributed systems, and it’s much more Python-native than Spark was. Ray also has much more of a focus on real-time performance. A lot of the things that people are interested in with Ray revolve around doing things like large-scale reinforcement learning—and it just so happens that deep reinforcement learning is something that everyone’s really excited about.
Related resources:
- “Optimization, compressed sensing, and large-scale machine learning pipelines”: the O’Reilly Data Show Podcast featuring Ben Recht.
- “Notes from the first Ray meetup”
- “Practical applications of reinforcement learning in industry”
- “Building tools for the AI applications of tomorrow”
- “Toward the Jet Age of machine learning”
In this episode of the Data Show, I spoke with Harish Doddi, co-founder and CEO of Datatron, a startup focused on helping companies deploy and manage machine learning models. As companies move from machine learning prototypes to products and services, tools and best practices for productionizing and managing models are just starting to emerge. Today’s data science and data engineering teams work with a variety of machine learning libraries, data ingestion, and data storage technologies. Risk and compliance considerations mean that the ability to reproduce machine learning workflows is essential to meet audits in certain application domains. And as data science and data engineering teams continue to expand, tools need to enable and facilitate collaboration.
As someone who specializes in helping teams turn machine learning prototypes into production-ready services, I wanted to hear what Doddi has learned while working with organizations that aspire to “become machine learning companies.”
Here are some highlights from our conversation:
A central platform for building, deploying, and managing machine learning models
In one of the companies where I worked, we had built infrastructure related to Spark. We were a heavy Spark shop. So we built everything around Spark and other components. But later, when that organization grew, a lot of people came from a TensorFlow background. That suddenly created a little bit of frustration in the team because everybody wanted to move to TensorFlow. But we had invested a lot of time, effort and energy in building the infrastructure for Spark.
… We suddenly had hidden technical debt that needed to be addressed. … Let’s say right now you have two models running in production and you know that in the next two or three years you are going to deploy 20 to 30 models. You need to start thinking about this ahead of time.
… That’s why these days I observed that organizations are creating centralized teams. The centralized team is responsible for maintaining flexible machine learning infrastructure that can be used to deploy, operate, and monitor many models simultaneously.
Feature store: Create, manage, and share canonical features
When I talk to companies these days, everybody knows that their data scientists are duplicating work because they don’t have a centralized feature store. Everybody I talk to really wants to build or even buy a feature store, depending on what is easiest for them.
… The number of data scientists within most companies is increasing. And one of the pain points I’ve observed is when a new data scientist joins an organization, there is an extreme amount of ramp-up period. A new data scientist needs to figure out what the data sets are, what the features are, so on and so forth. But if an organization had a feature store, the ramp-up period can be much faster.
Related resources:
- “Lessons learned turning machine learning models into real products and services”
- “What are machine learning engineers?”: examining a new role focused on creating data products and making data science work in production
- “MLflow: A Platform for Managing the Machine Learning Lifecycle”
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain
- “We need to build machine learning tools to augment machine learning engineers”
- When models go rogue: David Talby on hard-earned lessons about using machine learning in production
In this episode of the Data Show, I spoke with Chang Liu, applied research scientist at Georgian Partners. In a previous post, I highlighted early tools for privacy-preserving analytics, both for improving decision-making (business intelligence and analytics) and for enabling automation (machine learning). One of the tools I mentioned is an open source project for SQL-based analysis that adheres to state-of-the-art differential privacy (a formal guarantee that provides robust privacy assurances). Since business intelligence typically relies on SQL databases, this open source project is something many companies can already benefit from today.
What about machine learning? While I didn’t have space to point this out in my previous post, differential privacy has been an area of interest to many machine learning researchers. Most practicing data scientists aren’t aware of the research results, and popular data science tools haven’t incorporated differential privacy in meaningful ways (if at all). But things will change over the next months. For example, Liu wants to make ideas from differential privacy accessible to industrial data scientists, and she is part of a team building tools to make this happen.
Here are some highlights from our conversation:
Differential privacy and machine learning
In the literature, there are actually multiple ways differential privacy is used in machine learning. We can either inject noise directly at the input data level, or while we’re training a model. We can also inject noise into the gradient. At every iteration we’re computing the gradients, we can inject some sort of noise. Or we can also inject noise during aggregation. If we’re using ensembles, we can inject noise there. And we can also inject noise at the output level. So after we’ve trained the model, and we have our vectors of weights, then we can also inject noise directly to the weights.
A mechanism for building robust models
There could be a chance that differential privacy methods can actually make your model more general. Because, essentially, when models memorize their training data, it could be due to overfitting. So, injecting all of this noise may help the resulting model move you further away from overfitting, and you get a more general model.
Related resources:
- “How to build analytic products in an age when data privacy has become critical”
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain.
- “Data regulations and privacy discussions are still in the early stages”: Aurélie Pols on GDPR, ethics, and ePrivacy.
- “Data collection and data markets in the age of privacy and machine learning”
In this episode of the Data Show, I spoke with Andrew Feldman, founder and CEO of Cerebras Systems, a startup in the blossoming area of specialized hardware for machine learning. Since the release of AlexNet in 2012, we have seen an explosion in activity in machine learning, particularly in deep learning. A lot of the work to date happened primarily on general purpose hardware (CPU, GPU). But now that we’re six years into the resurgence in interest in machine learning and AI, these new workloads have attracted technologists and entrepreneurs who are building specialized hardware for both model training and inference, in the data center or on edge devices.
In fact, companies with enough volume have already begun building specialized processors for machine learning. But you have to either use specific cloud computing platforms or work at specific companies to have access to such hardware. A new wave of startups (including Cerebras) will make specialized hardware affordable and broadly available. Over the next 12-24 months architects and engineers will need to revisit their infrastructure and decide between general purpose or specialized hardware, and cloud or on-premise gear.
In light of the training duration and cost they face using current (general purpose) hardware, some experiments might be hard to justify. Upcoming specialized hardware will enable data scientists to try out ideas that they previously would have hesitated to pursue. This will surely lead to more research papers and interesting products as data scientists are able to run many more experiments (on even bigger models) and iterate faster.
As founder of one of the most anticipated hardware startups in the deep learning space, I wanted get Feldman’s views on the challenges and opportunities faced by engineers and entrepreneurs building hardware for machine learning workloads.
Here are some highlights from our conversation:
A renaissance for computer architecture
OpenAI put out some very interesting analysis recently that showed that since 2012, the compute use for the largest AI training runs has increased by 300,000x. … What’s available to us to attack the vast discrepancy between compute demand and what we have today? Two things are available to us. The first is, exploring interesting compute architectures. I think this ushers in a golden age for compute architectures. And number two, it’s building dedicated hardware and saying: ‘We’re prepared to make trade offs to accelerate AI compute by not trying to be good at other things. By not trying to be good at graphics or by not trying to be a good web server. But we will attack this vast demand for compute by building dedicated hardware for artificial intelligence work.’ Historically, the following has been a very productive and valuable trade off: new and interesting architectures dedicated for a particular type of work. That’s the opportunity that many of these hardware companies or chip companies have seen.
Communication intensive workloads
When you stay on a chip, one can communicate fairly quickly. The problem is our work in artificial intelligence often spans more than one traditional chip. And the performance penalty for leaving the chip is very, very high. On-chip, you stay in silicon; off-chip, you have to wrap your communication in some sort of protocol, you need to send it, connect it over lanes on a print circuit board or maybe through a PCI switch, or maybe through an Ethernet switch or an InfiniBand switch. This adds two, three, four orders of magnitude of latency.
Some of the problems the hardware vendors who are interested in solving data center training, data center inference are working on are how you can accelerate the communication between cores and across tens of thousands of cores, or even hundreds of thousands of cores across many chips. Some are inventing new techniques for special switches and modifying PCIe to do that. Others have sort of more fundamental approaches to accelerating this communication. But if you can’t communicate quickly, you can’t train a model quickly and you can’t provide inference quickly.
New hardware will significantly reduce training times
You should see a reduction in training times of 10-50x sometime over the next 12-18 months. … I think you’re going to see an additional 10-25x in the following year. We’re looking at three orders of magnitude in reduction in training time over the next several years.
Related resources:
- “How big compute is powering the deep learning rocket ship”: Greg Diamos on building computer systems for deep learning and AI.
- “The artificial intelligence computing stack”: A look at why the U.S. and China are investing heavily in this new computing stack
- “How to train and deploy deep learning at scale”: Ameet Talwalkar on large-scale machine learning.
- “Scaling machine learning”: Reza Zadeh on deep learning, hardware/software interfaces, and why.
- “A new benchmark suite for machine learning”: David Patterson and Gu-Yeon Wei on MLPerf, a new set of benchmarks compiled by industry and academic contributors.
- “Building tools for the AI applications of tomorrow”
- “Toward the Jet Age of machine learning”
In this episode of the Data Show, I spoke with Aurélie Pols of Mind Your Privacy, one of my go-to resources when it comes to data privacy and data ethics. This interview took place at Strata Data London, a couple of days before the EU General Data Protection Regulation (GDPR) took effect. I wanted her perspective on this landmark regulation, as well as her take on trends in data privacy and growing interest in ethics among data professionals.
Here are some highlights from our conversation:
GDPR is just the starting point
GDPR is not an end point. It’s a starting point for a journey where a balance between companies and society and users of data needs to be redefined. Because when I look at my children, I look at how they use technology, I look at how smart my house might become or my car or my fridge, I know that in the long run this idea of giving consent to my fridge to share data is not totally viable. What are we going to be build for the next generations?
… I’ve been teaching privacy and ethics in Madrid at the IE Business School, one of the top business schools in the world. I’ve been teaching in the big data and analytics graduate program. I see the evolution as well. Five years ago, they looked at me like, ‘What is she talking about?’ Three years ago, some of the people in the room started to understand. … Last year it was like ‘We get it.’
Privacy by design
It’s defined as data protection by design and by default as well. The easy part is more the default settings: when you create systems, it’s the question I ask 20 times a week: ‘Great. I love your system. What data do you collect by default and what do you pass on by default?’ Then you start turning things off and then we’ll see who takes on the responsibility to turn things on again. That’s a default. Privacy by design was pushed by Ann Cavoukian from Ottawa in Canada more than 10 years ago.
These principles are finding themselves within the legislation. Not only in GDPR—for example, Hong Kong is starting to talk about this and Japan as well. One of these principles is about positive-sum, not zero-sum. It’s not ‘I win and you lose.’ It’s ‘we work together and we both win.’ That’s a very good principle.
There are interesting challenges within privacy by design to translate these seven principles into technical requirements. I think there are opportunities as well. It talks about traceability, visibility, transparency. Which then comes back again to, we’re sitting on so much data; how much data do we want to surface and are data subjects or citizens ready to understand what we have, and are they able to make decisions based on that? … Hopefully this generation of more ethically minded engineers or data scientists will start thinking in that way as well.
Related resources:
- “The data subject first?”: Aurélie Pols draws a broad philosophical picture of the data ecosystem and then hones in on the right to data portability.
- “How to build analytic products in an age when data privacy has become critical”
- “Managing risk in machine learning models”: Andrew Burt and Steven Touw on how companies can manage models they cannot fully explain.
- “Building tools for the AI applications of tomorrow”
- “Toward the Jet Age of machine learning”
- “The real value of data requires a holistic view of the end-to-end data pipeline”: Ashok Srivastava on the emergence of machine learning and AI for enterprise applications
- “Bringing AI into the enterprise”: Kris Hammond on business applications of AI technologies and educating future AI specialists.
In this episode of the Data Show, I spoke with Andrew Burt, chief privacy officer at Immuta, and Steven Touw, co-founder and CTO of Immuta. Burt recently co-authored a white paper on managing risk in machine learning models, and I wanted to sit down with them to discuss some of the proposals they put forward to organizations that are deploying machine learning.
Some high-profile examples of models gone awry have raised awareness among companies for the need for better risk management tools and processes. There is now a growing interest in ethics among data scientists, specifically in tools for monitoring bias in machine learning models. In a previous post, I listed some of the key considerations organization should keep in mind as they move models to production, but the report co-authored by Burt goes far beyond and recommends lines of defense, including a description of key roles that are needed.
Here are some highlights from our conversation:
Privacy and compliance meet data science
Andrew Burt: I would say the big takeaway from our paper is that lawyers and compliance and privacy folks live in one world and data scientists live in another with competing objectives. And that can no longer be the case. They need to talk to each other. They need to have a shared process and some shared terminology so that everybody can communicate.
One of the recommendations that we make, taken from some of the model risk management frameworks, is to create what we call lines of defense: these are basically different lines of reviewers who conduct periodic reviews from the creation testing phase to validation to an auditing phase. And the members of those lines of review need to be made up of teams with multiple expertise. So, there needs to be data owners, the people responsible for the data being piped into the models; there needs to be compliance personnel who are thinking about legal and ethical obligations; there needs to be data scientists; and there needs to be subject domain experts.
… We also dive into how you should be thinking about de-risking and monitoring your input data. How you should be thinking about monitoring and de-risking your output data and using output data for models. … And then, I think really importantly, is this idea of thinking about what it means for a model to fail, and having a concrete plan for what that means, how to correct it if it fails, and how to pull it from production if you need to.
Explainability and GDPR
Steven Touw: I gave a talk, “How Does the GDPR Impact Machine Learning?”, at Strata Data London. A lot of people are concerned about language in GDPR, which states that you must be able to explain how the model came to that conclusion. I think people are kind of overreacting to this a little bit, and we need to inject some common sense in there along the lines of: at the end of the day, you can explain what data went in; you can explain the logic of what you’re trying to solve and why; and you don’t have to explain every neuron in the neural net and how it was correlated to every other piece.
I think the GDPR is actually doing a good thing. It’s enabling consumers to understand how the decisions are being made about them, but they don’t have to understand everything in the weeds about it. Because the whole point of machine learning is that it can do things that we can’t as humans. So, that’s why we use it, and there are cases where it makes sense to trust the model rather than humans to get things done and, potentially and hopefully, done more accurately.
Related resources:
- “How to build analytic products in an age when data privacy has become critical”
- “We need to build machine learning tools to augment machine learning engineers”
- “The ethics of artificial intelligence”
- “Interpreting predictive models with Skater: Unboxing model opacity”
- “Toward the Jet Age of machine learning”
- “Building tools for the AI applications of tomorrow”
- “The current state of applied data science”
In this episode of the Data Show, I spoke with Ashok Srivastava, senior vice president and chief data officer at Intuit. He has a strong science and engineering background, combined with years of applying machine learning and data science in industry. Prior to joining Intuit, he led the teams responsible for data and artificial intelligence products at Verizon. I wanted his perspective on a range of issues, including the role of the chief data officer, ethics in machine learning, and the emergence of AI technologies for enterprise products and applications.
Here are some highlights from our conversation:
Chief data officer
A chief data officer, in my opinion, is a person who thinks about the end-to-end process of obtaining data, data governance, and transforming that data for a useful purpose. His or her purview is relatively large. I view my purview at Intuit to be exactly that, thinking about the entire data pipeline, proper stewardship, proper governance principles, and proper application of data. I think that as the public learns more about the opportunities that can come from data, there’s a lot of excitement about the potential value that can be unlocked from it from the consumer standpoint, and also many businesses and scientific organizations are excited about the same thing. I think the CDO plays a role as a catalyst in making those things happen with the right principles applied.
I would say if you look back into history a little bit, you’ll find the need for the chief data officer started to come into play when people saw a huge amount of data coming in at high speeds with high variety and variability—but then also the opportunity to marry that data with real algorithms that can have a transformational property to them. While it’s true that CIOs, CTOs, and people who are in lines of business can and should think about this, it’s a complex enough process that I think it merits having a person and an organization think about that end-to-end pipeline.
Ethics
We’re actually right now in the process of launching a unified training program in data science that includes ethics as well as many other technical topics. I should say that I joined Intuit only about six months ago. They already had training programs happening worldwide in the area of data science and acquainting people with the principles necessary to use data properly as well as the technical aspects of doing it.
I really feel ethics is a critical area for those of us who work in the field to think about and to be advocates of proper use of data, proper use of privacy information and security, in order to make sure the data that we’re stewards of is used in the best possible way for the end consumer.
Describing AI
You can think about two overlapping circles. One circle is really an AI circle. The other is a machine learning circle. Many people think that that intersection is the totality of it, but in fact, it isn’t.
… I’m finding that AI needs to be bounded a little bit. I often say that it’s a reasonable technology with unreasonable expectations associated with it. I really feel this way, that people for whatever reason have decided that deep learning is going to solve many problems. And there’s a lot of evidence to support that, but frankly, there’s a lot of evidence also to support the fact that much more work has to be done before these things become “general purpose AI solutions.” That’s where a lot of exciting innovation is going to happen in the coming years.
Related resources:
- Understanding the Chief Data Officer: Lessons and advice from the role’s pioneers
- “Building tools for the AI applications of tomorrow”
- “Toward the Jet Age of machine learning”
- “How to build analytic products in an age when data privacy has become critical”
- “How big data and AI will reshape the automotive industry”: Evangelos Simoudis on next-generation mobility services
- “Bringing AI into the enterprise”: Kris Hammond on business applications of AI technologies and educating future AI specialists.
This episode of the Data Show marks our 100th episode. This podcast stemmed out of video interviews conducted at O’Reilly’s 2014 Foo Camp. We had a collection of friends who were key members of the data science and big data communities on hand and we decided to record short conversations with them. We originally conceived of using those initial conversations to be the basis of a regular series of video interviews. The logistics of studio interviews proved too complicated, but those Foo Camp conversations got us thinking about starting a podcast, and the Data Show was born.
To mark this milestone, my colleague Paco Nathan, co-chair of Jupytercon, turned the tables on me and asked me questions about previous Data Show episodes. In particular, we examined the evolution of key topics covered in this podcast: data science and machine learning, data engineering and architecture, AI, and the impact of each of these areas on businesses and companies. I’m proud of how this show has reached so many people across the world, and I’m looking forward to sharing more conversations in the future.
Here are some highlights from our conversation:
AI is more than machine learning
I think for many people machine learning is AI. I’m trying to, in the AI Conference series, convince people that a true AI system will involve many components, machine learning being one. Many of the guests I have seem to agree with that.
Evolving infrastructure for big data
In the early days of the podcast, many of the people I interacted with had Hadoop as one of the essential things in their infrastructure. I think while that might still be the case, there are more alternatives these days. I think a lot of people are going to object stores in the cloud. Another examples is that before, people maintained specialized systems. There’s still that, but people are trying to see if they can combine some of these systems, or come up with systems that can do more than one workload. For example, this whole notion in Spark of having a unified system that is able to do batch in streaming caught on during the span of this podcast.
Related resources:
- An easy to scan episode list of the Data Show
- “What is data science?”
- “What are machine learning engineers?”
- “What is Artificial Intelligence?”
- “Building tools for the AI applications of tomorrow”
- “Data engineering: A quick and simple definition”
In this episode of the Data Show, I spoke with Jason Dai, CTO of Big Data Technologies at Intel, and one of my co-chairs for the AI Conference in Beijing. I wanted to check in on the status of BigDL, specifically how companies have been using this deep learning library on top of Apache Spark, and discuss some newly added features. It turns out there are quite a number of companies already using BigDL in production, and we talked about some of the popular uses cases he’s encountered. We recorded this podcast while we were at the AI Conference in Beijing, so I wanted to get Dai’s thoughts on the adoption of AI technologies among Chinese companies and local/state government agencies.
Here are some highlights from our conversation:
BigDL: One year later
Big DL was actually first open-sourced on December 30, 2016—so it has been about 1 year and 4 months. We have gotten a lot of positive feedback from the open source community. We also added a lot of new optimizations and functionalities to Big DL. I think it roughly can be categorized into four classes. We did large optimizations, especially for the big data environment, which is essentially very large-scale Intel server clusters. We use a lot of hardware accelerations and Math Kernel libraries to improve BigDL’s performance on a single-node. At the same time, we leverage the Spark architecture so that we can efficiently scale out and perform very large-scale distributed training or inference.
The second part of the year we provided very rich support for existing deep learning tools. We can actually directly load and save models to and from TensorFlow, Caffe, Keras, Torch, and other libraries. People can take the existing models and actually load into Big DL and run it on Spark.
End-to-end machine learning pipelines End-to-end object detection and image feature extraction pipeline on top of Spark and BigDL. Image source: BigDL white paper, used with permission.
An industrial deep learning or machine learning application is actually a very complex end-to-end, big data analytics pipeline. You start with the data ingestion, data processing, ETL, and then after that, you will transform your data. For instance, you can do image augmentations, text tokenization, word embedding, and so on. After that, you will extract the features or perform various feature transformations and feature extractions. After that, when you have the feature extracted and transformed, then you will probably begin with your model training. But even the model training itself could be an iterative process, a pipeline, if you want to introduce various hyperparameter tunings.
…There’s actually a reason why we took an integrated approach when we built Big DL on top of Apache Spark: you know that you’re going back and forth between data ingestion, data processing, model training, and inference. Having integrated software and hardware infrastructure will benefit the user a lot.
AI in China
People and companies in China have a very high level of awareness and a high level of hope that AI can be used to solve real problems. In China, you have the advantage in that people move fast to apply these new AI technologies. Companies here have access to large amounts of data, and there are many use cases across industries and the public sector. There are a lot of ways you can actually apply new technology and AI to real-world applications and see their impact.
In general, people and companies in China move very fast, and they are very good at experimenting and trying new approaches. People here think they can always iterate and refine their approaches, and we see a lot of new technology getting put into practice very quickly.
Related resources:
- “What happens when AI experts from Silicon Valley and China meet”
- BigDL: Jason Dai on the launch of a new deep learning library for Apache Spark
- “Why AI and machine learning researchers are beginning to embrace PyTorch”: Soumith Chintala on building a worthy successor to Torch and on deep learning within Facebook.
- MXNet: Anima Anandkumar on deep learning that’s easy to implement and easy to scale
- “Introducing RLlib—A composable and scalable reinforcement learning library”: this new software makes the task of training RL models much more accessible
- “How machine learning will accelerate data management systems”: Tim Kraska on why ML will change how we build core algorithms and data
In this episode of the Data Show, I spoke with Jerry Overton, senior principal and distinguished technologist at DXC Technology. I wanted the perspective of someone who works across industries and with a variety of companies. I specifically wanted to explore the current state of data science and AI within companies and public sector agencies. As much as we talk about use cases, technologies, and algorithms, there are also important issues that practitioners like Overton need to address, including privacy, security, and ethics. Overton has long been involved in teaching and mentoring new data scientists, so we also discussed some tips and best practices he shares with new members of his team.
Here are some highlights from our conversation:
Where most companies are in their data journey
Five years ago, we had this moneyball phase where moneyball was new. This idea that you could actually get to value with data, and that data would have something to say that could help you run your business better.
We’ve gone way past that now to where I think it’s pretty much a premise that if you aren’t using your data, you’re losing out on a very big competitive advantage. I think it’s pretty much a premise that data science is necessary and that you need to do something. Now, the big thing is that companies are really unsure as to what their data scientists should be doing—which areas of their business they can make smarter and how to make it smarter.
… Then, you add artificial intelligence on top of this. Companies hear a lot about artificial intelligence, and they have seen some pretty cool demos—what you can do with extending domain expertise or complex planning, inferring intent, things like that. We’re entering that same phase where there are a lot of companies that are kind of skeptical as to whether or not it can actually help them.
Enterprise data science. Image by Jerry Overton, used with permission. Ethics, fairness, and transparency in analytics
Here are some standard things that we bring to projects. First, we have to build forensic tools to profile the algorithms being used, and then after we have a profile, we anticipate its behavior. Then we get together a diverse group and we assess the enterprise risk.
… Getting people to understand that you have to handle ethics, and fairness, and bias—there are usually pretty mature programs in place for doing that. But where companies have problems is in the specific tactics for doing that. Strategies for making sure that what you put out is aligned with the ethics of the group of the company that you work in, and that’s a lot of what we help our clients with.
Related resources:
- “Going Pro in Data Science”: Jerry Overton on what it takes to succeed as a professional data scientist
- “Bringing AI into the enterprise”: Kris Hammond on business applications of AI technologies and educating future AI specialists
- “We need to build machine learning tools to augment machine learning engineers”
- “Machine learning at Spotify”: Christine Hung on using data to drive digital transformation and recommenders that increase user engagement
- “Transforming organizations through analytics centers of excellence”: Carme Artigas on helping enterprises transform themselves with big data tools and technologies
In this episode of the Data Show, I spoke with Guillaume Chaslot, an ex-YouTube engineer and founder of AlgoTransparency, an organization dedicated to helping the public understand the profound impact algorithms have on our lives. We live in an age when many of our interactions with companies and services are governed by algorithms. At a time when their impact continues to grow, there are many settings where these algorithms are far from transparent. There is growing awareness about the vast amounts of data companies are collecting on their users and customers, and people are starting to demand control over their data. A similar conversation is starting to happen about algorithms—users are wanting more control over what these models optimize for and an understanding of how they work.
I first came across Chaslot through a series of articles about the power and impact of YouTube on politics and society. Many of the articles I read relied on data and analysis supplied by Chaslot. We talked about his work trying to decipher how YouTube’s recommendation system works, filter bubbles, transparency in machine learning, and data privacy.
Here are some highlights from our conversation:
Why YouTube’s impact is less understood
My theory why people completely overlooked YouTube is because on Facebook and Twitter, if one of your friends posts something strange, you’ll see it. Even if you have 1,000 friends, if one of them posts something really disturbing, you see it, so you’re more aware of the problem. Whereas on YouTube, some people binge watch some very weird things that could be propaganda, but we won’t know about it because we don’t see what other people see. So, YouTube is like a TV channel that doesn’t show the same thing to everybody and when you ask YouTube, “What did you show to other people?” YouTube says, ‘I don’t know, I don’t remember, I don’t want to tell you.’
Downsides of optimizing only for watch time
When I was working on the YouTube algorithm and our goal was to optimize watch time, we were trying to make sure that the algorithm kept people online the longest. But what I realized was that we were so focused on this target of watch time that we were forgetting a lot of important things and we were seeing some very strange behavior of the algorithm. Each time we were seeing this strange behavior, we just blamed it on the user. It shows violent videos; it must be because users are violent, so it’s not our fault; the algorithm is just a mirror of human society. But if I believe the algorithm is a mirror of human society, I think it’s also not a flat mirror; it’s a mirror that emphasizes some aspects of life and makes some other aspects overlooked.
… The algorithm that is behind YouTube and the Facebook news feeds are very complex, deep learning systems that will take a lot into account, including user sessions, what they’ve watched. It will try to find the right content to show to users to get them to stay online the longest and interact as much as possible with the content. So, this can seem neutral at first, but it might not be neutral. For instance, if you have content that says ‘The media is lying,’ whether it’s on Facebook or on YouTube, what will happen is that this content will naturally, if it manages to convince the user that the media is lying, the content will be very efficient at keeping the user online because the user won’t go to other media and will spend more time on YouTube and more time on Facebook.
… In my personal opinion, the current goal of maximizing watch time means that any content that is really good at captivating your attention for a long time will perform really well. This means extreme content will actually perform really well. But say you had another goal—for instance, the goal to maximize likes and dislikes, or another system of rating like when you would be asked some question like, ‘Did you enjoy this video? Was it helping you in your life?’ Then this kind of extreme content will not perform as well. So, there are many other options, and it’s not that YouTube is failing at exploring these other options; it’s that they don’t even try.
Related resources:
- “We need to build machine learning tools to augment machine learning engineers”
- “Defining responsible data practices”: Natalie Evans Harris discusses the community principles on ethical data practices (CPEDP), a code of ethics for data collection, sharing, and utilization.
- “Inclusivity for the greater good”: Ajey Gore explains why GO-JEK is focusing its attention beyond urban Indonesia to help people across the country’s rural areas.
- “Ethics in data project design: It’s about planning”
- “On computational ethics”
- Ethics of Big Data: Doug Patterson and Kord Davis on balancing risk and innovation.
In this episode of the Data Show, I spoke with Jesse Anderson, managing director of the Big Data Institute, and my colleague Paco Nathan, who recently became co-chair of Jupytercon. This conversation grew out of a recent email thread the three of us had on machine learning engineers, a new job role that LinkedIn recently pegged as the fastest growing job in the U.S. In our email discussion, there was some disagreement on whether such a specialized job role/title was needed in the first place. As Eric Colson pointed out in his beautiful keynote at Strata Data San Jose, when done too soon, creating specialized roles can slow down your data team.
We recorded this conversation at Strata San Jose, while Anderson was in the middle of teaching his very popular two-day training course on real-time systems. We closed the conversation with Anderson’s take on Apache Pulsar, a very impressive new messaging system that is starting to gain fans among data engineers.
Here are some highlights from our conversation:
Why we need machine learning engineers
Jesse Anderson: (2:09) One of the issues I’m seeing as I work with teams is that they’re trying to operationalize machine learning models, and the data scientists are not the one to productionize these. They simply don’t have the engineering skills. Conversely, the data engineers don’t have the skills to operationalize this either. So, we’re seeing this kind of gap in between the data science and the data engineering, and the gap I’m seeing and the way I’m seeing it being filled, is through a machine learning engineer.
… I disagree with Paco that generalization is the way to go. I think it’s hyper-specialization, actually. This is coming from my experience having taught a lot of enterprises. At a startup, I would say that super-specialization is probably not going to be as possible, but at an enterprise, you are going to have to have a team that specializes in big data, and that is a part from a team, even a software engineering team, that doesn’t work with data.
Putting Apache Pulsar on the radar of data engineers Key features of Apache Pulsar. Image by Karthik Ramasamy, used with permission.
Jesse Anderson: (23:30) A lot of my time, since I’m really teaching data engineering is spent on data integration and data ingestion. How do we move this data around efficiently? For a lot of that time Kafka was really the only open source game in town for that. But now there’s another technology called Apache Pulsar. I’ve spent a decent amount of time actually going through Pulsar and there are some things that I see in it that Kafka will either have difficulty doing or won’t be able to do.
… Apache Pulsar separates pub-sub from storage. When I first read about that, I didn’t quite get it. I didn’t quite see, why is this so important or why this is so interesting. It’s because you can individually scale your pub-sub and your storage resources independently. Now you’ve got something. Now you can say, “Well, we originally decided I wanted to store data for seven days. All right, let’s spin up some more bookkeeper processes and now we can store fourteen days, now we can store twenty one days.” I think that’s going to be a pretty interesting addition there. Where the other side of that, the corollary to that is, “Okay, we’re hitting Black Friday and we don’t have so much more data coming through as we have way more consumption and have way more things hitting our pub-sub. We could spin up more pub-sub with that.” This separation is actually allowing some interesting use cases.
Related resources:
- “What are machine learning engineers?”
- “We need to build machine learning tools to augment machine learning engineers”
- “Differentiating via data science”: Eric Colson explains why companies must now think very differently about the role and placement of data science in organizations.
- “Architecting and building end-to-end streaming applications”: Karthik Ramasamy on Heron, DistributedLog, and designing real-time applications.
- “How machine learning will accelerate data management systems”: Tim Kraska on why ML will change how we build core algorithms and data
In this episode of the Data Show, I spoke with Ameet Talwalkar, assistant professor of machine learning at CMU and co-founder of Determined AI. He was an early and key contributor to Spark MLlib and a member of AMPLab. Most recently, he helped conceive and organize the first edition of SysML, a new academic conference at the intersection of systems and machine learning (ML).
We discussed using and deploying deep learning at scale. This is an empirical era for machine learning, and, as I noted in an earlier article, as successful as deep learning has been, our level of understanding of why it works so well is still lacking. In practice, machine learning engineers need to explore and experiment using different architectures and hyperparameters before they settle on a model that works for their specific use case. Training a single model usually involves big (labeled) data and big models; as such, exploring the space of possible model architectures and parameters can take days, weeks, or even months. Talwalkar has spent the last few years grappling with this problem as an academic researcher and as an entrepreneur. In this episode, he describes some of his related work on hyperparameter tuning, systems, and more.
Here are some highlights from our conversation:
Deep learning
I would say that you hear a lot about the modeling of problems associated with deep learning. How do I frame my problem as a machine learning problem? How do I pick my architecture? How do I debug things when things go wrong? … What we’ve seen in practice is that, maybe somewhat surprisingly, the biggest challenges that ML engineers face actually are due to the lack of tools and software for deep learning. These problems are sort of like hybrid systems/ML problems. Very similar to the sorts of research that came out of the AMPLab.
… Things like TensorFlow and Keras, and a lot of those other platforms that you mentioned, are great and they’re a great step forward. They’re really good at abstracting low-level details of a particular learning architecture. In five lines, you can describe how your architecture looks and then you can also specify what algorithms you want to use for training.
There are a lot of other systems challenges associated with actually going end to end, from data to a deployed model. The existing software solutions don’t really tackle a big set of these challenges. For example, regardless of the software you’re using, it takes days to weeks to train a deep learning model. There’s real open challenges of how to best use parallel and distributed computing both to train a particular model and in the context of tuning hyperparameters of different models.
We also found out the vast majority of organizations that we’ve spoken to in the last year or so who are using deep learning for what I’d call mission-critical problems, are actually doing it with on-premise hardware. Managing this hardware is a huge challenge and something that folks like me, if I’m working at a company with machine learning engineers, have to figure out for themselves. It’s kind of a mismatch between their interests and their skills, but it’s something they have to take care of.
Understanding distributed training
To give a little bit more background, the idea behind this work started about four years ago. There was no deep learning in Spark MLlib at the time. We were trying to figure out how to perform distributed training of deep learning in Spark. Before actually getting our hands really dirty and trying to actually implement anything we wanted to just do some back-of-the-envelope calculations to see what speed-ups you could hope to get.
… The two main ingredients here are just computation and communication. … We wanted to understand this landscape of distributed training, and, using Paleo, we’ve been able to get a good sense of this landscape without actually running experiments. The intuition is simple. The idea is that if we’re very careful in our bookkeeping, we can write down the full set of computational operations that are required for a particular neural network architecture when it’s performing training.
[Full disclosure: I’m an advisor to Determined AI.]
Related resources:
- “Introducing RLlib—A composable and scalable reinforcement learning library”: this new software makes the task of training RL models much more accessible
- “Scaling machine learning”: Reza Zadeh on deep learning, hardware/software interfaces, and why computer vision is so exciting
- “How big compute is powering the deep learning rocket ship”: Greg Diamos on building computer systems for deep learning and AI.
- “We need to build machine learning tools to augment machine learning engineers”
- “How machine learning will accelerate data management systems”: Tim Kraska on why ML will change how we build core algorithms and data
In this episode of the Data Show, I spoke with Ofer Ronen, GM of Chatbase, a startup housed within Google’s Area 120. With tools for building chatbots becoming accessible, conversational interfaces are becoming more prevalent. As Ronen highlights in our conversation, chatbots are already enabling companies to automate many routine tasks (mainly in customer interaction). We are still in the early days of chatbots, but if current trends persist, we’ll see bots deployed more widely and take on more complex tasks and interactions. Gartner recently predicted that by 2021, companies will spend more on bots and chatbots than mobile app development.
Like any other software application, as bots get deployed in real-world applications, companies will need tools to monitor their performance. For a single, simple chatbot, one can imagine developers manually monitoring log files for errors and problems. Things get harder as you scale to more bots and as the bots get increasingly more complex. As in the case of other machine learning applications, when companies start deploying many more chatbots, automated tools for monitoring and diagnostics become essential.
The good news is relevant tools are beginning to emerge. In this episode, Ronen describes a tool he helped build: Chatbase is a chatbot analytics and optimization service that leverages machine learning research and technologies developed at Google. In essence, Chatbase lets companies focus on building and deploying the best possible chatbots.
Here are some highlights from our conversation:
Democratization of tools for bot developers
It’s been hard to get the natural language processing to work well and to recognize all the different ways people might say the same thing. There’s been an explosion of tools that leverage machine learning and natural language processing (NLP) engines to make sense of all that’s being asked of bots. But with increased capacity and capability to process data, there’s now better third-party tools for any company to take advantage of and build a decent bot out of the box.
… I see three levels of bot builders out there. There’s the non-technical kind where marketing or sales might create a prototype using a user interface—like maybe Chatfuel, which requires no programming, and create a basic experience. Or they might even create some sort of decision tree bot that is not flexible, but is good for maybe basic lead-gen experiences. But they often can’t handle type-ins. It’s often button-based. So, that’s one level, the non-technical folks.
Then there are teams that have developers on staff. They’re not machine learning experts, but they’re developers that can use off-the-shelf natural language processing engines to extract meaning from messages sent by users. So, you’re extracting intents and entities and making sense of what’s coming at your bot without having to have machine learning expertise.
Finally, there are teams that have the machine learning experts. They might build their own NLP engine to give them more control over how it works. But often, that’s not needed if a third-party solution can serve most of your needs. But we do see some teams like that.
Popular use cases
We track tens and tens of thousands of bots each month with Chatbase, and what we see is that for large companies, they often start with customer support for at least two reasons. One is that the automation can save them some money, but also because chatbots enable them to create a more effective experience for their users. In fact, there’s a survey by Salesforce and a couple other companies that found that what people want from bots is quick 24/7 answers to simple questions.
We also see some lead-generation bots. Those are simpler to build and often just live on a website of a company to try to gather and qualify leads. They can, most of the time, do a decent job and do a little better than just a dumb form. They can be responsive as they collect information from the user. They provide a more interactive experience, and, in many cases, can be more delightful for the customer.
Common errors: Unsupported, misunderstood, missed
Unsupported requests are ones that you just haven’t built, like someone is asking a travel bot to upgrade their seat, and you just haven’t built that functionality. As a product manager, let’s say, running a bot project, you’d want to know the top-requested features. You can use machine learning to cluster the similar requests and show the product what to build next based on popularity.
Misunderstood is when the user asks for X, but the bot thinks they asked for Y. That can be a really annoying experience for the end user. You’re likely going to lose the end user in these situations. These happen more often when a bot is expanding its use cases, which are referred to as intent and handles. The more intents and handles, the more of an opportunity to mislabel and misunderstand the user. Chatbase looks for these mislabeled or misunderstood situations, and we suggest alternate intents. We suggest fixes.
The final one is missed requests. This is when you’ve built the functionality, you have an intent for a certain type of request, and maybe you recognize a hundred ways of asking for that specific intent. There’s maybe another hundred that you don’t recognize. So, instead of having, let’s say, a business analyst go through your logs to find the other hundred, we automatically detect these extra ways of saying the same thing and suggest those to you so you can fix the bot.
Note: this episode of the Data Show is a collaboration between O’Reilly and Chatbase. See our statement of editorial independence.
Related resources:
- “Bots: What you need to know”
- “An overview of the bot landscape”
- “Commercial speech recognition systems in the age of big data and deep learning”: Yishay Carmiel on applications of deep learning in text and speech
- “We need to build machine learning tools to augment machine learning engineers”
- “Topic models—past, present, and future”: David Blei discusses the origins and applications of topic models.
In this episode of the Data Show, I spoke with Danny Lange, VP of AI and machine learning at Unity Technologies. Lange previously led data and machine learning teams at Microsoft, Amazon, and Uber, where his teams were responsible for building data science tools used by other developers and analysts within those companies. When I first heard that he was moving to Unity, I was curious as to why he decided to join a company whose core product targets game developers.
As you’ll glean from our conversation, Unity is at the forefront of some of the most exciting, practical applications of deep learning (DL) and reinforcement learning (RL). Realistic scenery and imagery are critical for modern games. GANs and related semi-supervised techniques can ease content creation by enabling artists to produce realistic images much more quickly. In a previous post, Lange described how reinforcement learning opens up the possibility of training/learning rather than programming in game development.
Lange explains why simulation environments are going to be important tools for AI developers. We are still in the early days of machine intelligence, and I am looking forward to more tools that can democratize AI research (including future releases by Lange and his team at Unity).
Here are some highlights from our conversation:
Why reinforcement learning is so exciting
I’m a huge fan of reinforcement learning. I think it has incredible potential, not just in game development but in a lot of other areas, too. … What we are doing at Unity is basically making reinforcement learning available to the masses. We have shipped open source software on GitHub called Unity ML Agents, that include the basic frameworks for people to experiment with reinforcement learning. Reinforcement learning is really creating a machine learned-driven feedback loop. Recall the example I previously wrote about, of the chicken crossing the road; yes, it gets hit thousands and thousands of times by these cars, but every time it gets hit, it learns that’s a bad thing. And every time it manages to pick up a gift package on the way over the road, that’s a good thing.
Over time, it gets superhuman capabilities in crossing this road, and that is fantastic because there’s not a single line of code going into that. It’s pure simulation, and through reinforcement learning it captures a method. It learns a method to cross the road, and you can take that into many different aspects of games. There are many different methods you can train. You can add two chickens—can they collaborate to do something together? We are looking at what we call multi-agent systems, where two or more of these trained reinforcement learning-trained agents are acting together to achieve a goal.
… I want a million developers to start working on this. I want a lot more innovation, and I want a lot more out-of-the-box thinking, and that is what we want by making our RL tools and platform available to our Unity community. Let me just jump to one thing here: most people think that reinforcement learning in the game world or in game-like situations is a lot about what we call ‘path finding.’ Path finding is basically for a character in a game to navigate through some situation—this is pretty well understood. There are good algorithms for that. Looking ahead, I’m actually thinking about a different set of decisions. For instance, which weapon or which tool should a character pick up and bring with them in a game? That is a much, much harder decision. It’s strategy at a higher level.
Machine learning and AI at Unity
If you think about where intelligence originated around us (animals and humans), it’s really originating out of surviving and thriving in a physical world. That is really the job of intelligence. You have to survive, you have to find food, you have to avoid your enemies, you have to walk falling down—so, gravity is playing a big role there. If you think about the Unity game engine, it creates a visual 3D environment where the laws of physics rule. So, you have gravity, you have inertia, you have friction, and you basically have a 3D environment.
Unity provides a fantastic lab to explore machine learning, and thereby elements of artificial intelligence in simulations within that world. So, rather than just using machine learning to work on spreadsheets and sell more products at Amazon or get your car to arrive more quickly at Uber, now you can actually start running simulations that are covering aspects of the real world, and you can explore things like vision, touching, path finding, etc.
Related resources:
- ”Practical applications of reinforcement learning in industry”
- “Bringing gaming to life with AI and deep learning”: Danny Lange on how machine learning opens the door for the use of training rather than programming in game development
- “Introducing RLlib – A composable and scalable reinforcement learning library”: this new software makes the task of training RL models much more accessible
- “An Outsider’s Tour of Reinforcement Learning”: Ben Recht’s ongoing series of posts
- “Open-endedness: The last grand challenge you’ve never heard of”
- “Why continuous learning is key to AI”
In this episode of the Data Show, I spoke with Leo Meyerovich, co-founder and CEO of Graphistry. Graphs have always been part of the big data revolution (think of the large graphs generated by the early social media startups). In recent months, I’ve come across companies releasing and using new tools for creating, storing, and (most importantly) analyzing large graphs. There are many problems and use cases that lend themselves naturally to graphs, and recent advances in hardware and software building blocks have made large-scale analytics possible.
Starting with his work as a graduate student at UC Berkeley, Meyerovich has pioneered the combination of hardware and software acceleration to create truly interactive environments for visualizing large amounts of data. Graphistry has built a suite of tools that enables analysts to wade through large data sets and investigate business and security incidents. The company is currently focused on the security domain—where it turns out that graph representations of data are things security analysts are quite familiar with.
Here are some highlights from our conversation:
Graphs as the front end for machine learning
They’re really flexible. First of all, there’s a pure analytic reason in that there are certain types of queries that one could do efficiently with a graph database. If you needed do a bunch of joins, graphs are really great at that. … Companies want to get into stuff like 360-degree views of things; they want to understand correlations to actually explain what’s going on at a more intelligent level.
… I think that’s where graphs really start to shine. Because companies deal with pretty heterogeneous data, and a graph ends up being a really easy way to deal with that. A lot of questions are basically, “What’s nearby?”—almost like your nearest neighbor type of stuff; the graph becomes, both at the query level and at the visual level, very interpretable. I now have a hypothesis about graphs as being the front end and the UI for machine learning, but that might be a topic for another day.
Graph applications and correlation services
If we’re talking about investigating a financial crime, you’ve got a transaction or user. … For example, if the user has multiple names but all the names are using the same address, you’re going to want to see that relationship.
… In security, where a lot of my mind is today, there is something called Kill Chain, where if you think of any bad incident, there’s probably a sequence of events around it, that led up to it. … You can map out that Kill Chain. So, in a sense, a lot of the reason Graphistry uses graphs is so we can let people see that sort of progression of events and reason about it.
… When people are using the graphs, especially in an enterprise setting, I think there’s a process change that’s happening if you’re building an enterprise data lake type of system. … It’s great if you can get individual alerts and create cases and investigations around individual alerts. But increasingly, you want a higher-level thing. … Instead of looking at individual alerts or individual events, you really want to think of incidents—an incident is basically a collection of alerts. For example, maybe there’s some fraud going on; if somebody figured out how to do fraud once, they’re probably going to try doing it multiple times. So, you don’t want to be playing whack-a-mole on little symptoms in each individual case; you want to get that full group incident. A graph becomes basically a way to create a real correlation service.
(Full disclosure: I’m an advisor to Graphistry.)
Related resources:
- “Graph databases are powering mission-critical applications”: Emil Eifrem on popular applications of graph technologies
- “Semi-supervised, unsupervised, and adaptive algorithms for large-scale time series”: Ira Cohen on developing machine learning tools for a broad range of real-time applications
- “Using Apache Spark to predict attack vectors among billions of users and trillions of events“: Fang Yu on data science in security
- “How machine learning will accelerate data management systems”
- “Architecting and building end-to-end streaming applications”
In this episode of the Data Show, I spoke with Mark Hammond, founder and CEO of Bonsai, a startup at the forefront of developing AI systems in industrial settings. While many articles have been written about developments in computer vision, speech recognition, and autonomous vehicles, I’m particularly excited about near-term applications of AI to manufacturing, robotics, and industrial automation. In a recent post, I outlined practical applications of reinforcement learning (RL)—a type of machine learning now being used in AI systems. In particular, I described how companies like Bonsai are applying RL to manufacturing and industrial automation. As researchers explore new approaches for solving RL problems, I expect many of the first applications to be in industrial automation.
Here are some highlights from our conversation:
Machine learning and machine teaching
Everyone is so focused on making better and faster learning algorithms; what do we do when we have it? Let’s just suppose that you now have an algorithm that can learn as well as or better than humans. How do we use that, how do we apply that in a predictable, scalable, repeatable way toward the objectives that we want to apply it toward?
… I thought about that for a while, and it’s one of those things where the answer is obvious in hindsight, but until you sit down and really chew on it, it doesn’t jump out at you. And it’s that, by design, if you’re building a learning system—if you want to program it—you have to teach it. Machine teaching and machine learning are necessary complements to one another; you need both. And for the large part, most of what comprises machine teaching these days consists of giant label data sets.
… You need machine teaching and machine learning. It dawned on me that this was the core abstraction that was going to make it possible for us to start applying all of this stuff more broadly across all the myriad use cases that we see in the real world without having to turn all of the people who are looking to use it into experts in machine learning and data science. It’s what enabled me to realize what Bonsai’s mission is: to enable your subject matter experts (a chemical engineer or a mechanical engineer, someone who is very, very well versed in whatever their domain is but not necessarily in machine learning or data science) to take that expertise and use it as the foundation for describing what to teach and then automating the underlying pieces for how you can actually effectively learn that.
Common use cases for reinforcement learning
In a lot of cases, the system you are working with is undergoing some form of tuning. We see a lot of this. It might be that you have an HVAC system and you are looking to optimize across the lifetime of the equipment, the comfort of the user, and the energy consumption of the device, which is its own complex optimization problem. You do this by virtue of tuning a bunch of parameters or controlling a bunch of different capabilities on the device.
… So, in many cases, a lot of what our users are applying the system toward is not necessarily strict real-time control. Certainly, we do that as well, and we put out a paper recently showing dexterous robotic control and manipulation using these kinds of techniques. You can absolutely do that, and we have customers who do that. A very common thing that we see across a lot of the real-world applications is more a tuning characteristic.
Related resources:
- “Practical applications of reinforcement learning in industry”: an overview of commercial and industrial applications of reinforcement learning
- “Introducing RLlib – A composable and scalable reinforcement learning library”: this new software makes the task of training RL models much more accessible
- Deep reinforcement learning in the enterprise—Bridging the gap from games to industry (2017 Artificial Intelligence Conference presentation by Mark Hammond)
- Ray: A distributed execution framework for reinforcement learning applications (2017 Artificial Intelligence Conference presentation by Ion Stoica)
- Deep reinforcement learning for robotics (2016 Artificial Intelligence Conference presentation by Pieter Abbeel)
- “Neuroevolution: A different kind of deep learning”
In this episode of the Data Show, I spoke with Fabian Yamaguchi, chief scientist at ShiftLeft. His 2015 Ph.D. dissertation sketched out how the combination of static analysis, graph mining, and machine learning, can be used to develop tools to augment security analysts. In a recent post, I argued for machine learning tools to augment teams responsible for deploying and managing models in production (machine learning engineers). These are part of a general trend of using machine learning to develop and manage the software systems of tomorrow. Yamaguchi’s work is step one in this direction: using machine learning to reduce the number of security vulnerabilities in complex software products.
Here are some highlights from our conversation:
Machine learning to find code vulnerabilities
I was not trying to build something that would just automatically take the code and give you all of the vulnerabilities. Instead, I was looking at the typical kind of tasks that I would encounter myself when doing these security audits, and I would ask myself, how can I automate these subtasks? As an example, when you find a vulnerability in code, the question that often arises is whether there are similar vulnerabilities still in that same program. That’s one of those subtasks you can automate well because what you’re actually doing is saying: ‘Hey, here’s an example of what a bug looks like. Can you scan the rest of the code? Can you use machine learning to actually determine other locations in the code that implement the same bug?’
… In machine learning, you never have enough data. In this case, this is actually an unsupervised learning approach. You’re taking all of the functions that you can get and you extract the dominant programming patterns in there. … It’s a bit like what you would do to find similar text documents, but it’s used for code.
From source code to graph analytics
By transforming software code into a graph, you can actually extract different properties from that code by analyzing the graph.
… Let’s take a smaller function that might have one IF block. One of the graph structures that’s first generated is called an abstract syntax tree. That’s a tree that you’d get by just parsing the code. … For each IF and for each variable, for each statement, there’s going to be a node. For each operator, like if there’s an assignment, there’s also going to be a node, and they are all connected by edges. You soon run into a lot of nodes and edges. If you take something like, let’s say, the Linux kernel, you’ll have several hundreds of thousands of nodes.
… You can do a lot by essentially solving reachability problems in these graphs.
Related resources:
- “How machine learning will accelerate data management systems”
- Artificial intelligence in the software engineering workflow: A 2017 AI Conference keynote by Peter Norvig
- “Responsible deployment of machine learning”: Why we need to build machine learning tools to augment our machine learning engineers
- “Architecting and building end-to-end streaming applications”
- Data is only as valuable as the decisions it enables
In this episode of the Data Show, I spoke with Kristian Hammond, chief scientist of Narrative Science and professor of EECS at Northwestern University. He has been at the forefront of helping companies understand the power, limitations, and disruptive potential of AI technologies and tools. In a previous post on machine learning, I listed types of uses cases (a taxonomy) for machine learning that could just as well apply to enterprise applications of AI. But how do you identify good use cases to begin with?
A good place to start for most companies is by looking for AI technologies that can help automate routine tasks, particularly low-skill tasks that occupy the time of high-skilled workers. An initial list of candidate tasks can be gathered by applying the following series of simple questions:
- Is the task data-driven?
- Do you have the data to support the automation of the task?
- Do you really need the scale that automation can provide?
We discussed other factors companies should consider when thinking through their AI strategies, education and training programs for AI specialists, and the importance of ethics and fairness in AI and data science.
Here are some highlights from our conversation:
It begins with finding use cases
I’ve been interacting more and more with the companies that are thinking about AI solutions; they often won’t have gotten to the place where they can talk about what they want to do. It’s an odd thing because there’s so much data out there and there’s so much hunger to derive something from that data. The starting point is often bringing an organization back down to, “So what do you want and need to do? What kind of decision-making do you want to support? What kinds of predictions would you like to be able to make?”
Identifying which tasks can be automated
Sometimes, you see a decision being made and, from an organizational point of view, everyone agrees that this decision is really strongly data driven. But it’s not strongly data driven. It’s data driven based upon the historical information that two or three people are using. It looks like they’re looking at data and then making a decision, but, in fact, what they’re doing is, they’re looking at data and they’re remembering one of 2,000 past examples in their heads and coming out with a decision.
… There are sets of tasks in almost any organization that nobody likes to have anything to do with. In the legal profession, there are tasks around things like discovery where you actually need to be able to look through a corpus of documents, but you need to have also some idea of the semantic relationships between words. This is totally learnable using existing technologies.
… It’s not as though tasks that can be automated don’t exist. They do, and, in fact, they not only exist, but they’re easily doable with current technologies. It’s a matter of understanding where to draw the line. It’s sometimes easy for organizations to look at the problem and sort of hallucinate that there is not a different kind of reasoning going on in the heads of the people who are solving the problem.
… You have to be willing to look at that and say, “Oh, I’m not going to replace the smartest person in the company, but, you know, I will free up the time of some of our smartest people by taking these tasks on and having the machine do them.”
Related resources:
- Here and now – Bringing AI into the enterprise: Kris Hammond’s tutorial at the 2017 AI conference in San Francisco.
- Vertical AI – Solving full stack industry problems using subject-matter expertise, unique data, and AI to deliver a product’s core value proposition: Bradford Cross at the 2017 AI conference in San Francisco.
- Demystifying the AI hype: Kathryn Hume at the 2017 AI conference in NYC.
- “6 practical guidelines for implementing conversational AI“: Susan Etlinger on how organizations can create more fluid interactions between humans and machines.
- “How companies can navigate the age of machine learning“: to become a “machine learning company,” you need tools and processes to overcome challenges in data, engineering, and models.
In this episode of the Data Show, I spoke with Tim Kraska, associate professor of computer science at MIT. To take advantage of big data, we need scalable, fast, and efficient data management systems. Database administrators and users often find themselves tasked with building index structures (“indexes” in database parlance), which are needed to speed up data access.
Some common examples include:
- B-Trees—used for range requests (e.g., assemble all sales orders within a certain time frame)
- Hash maps—used for key-based lookups
- Bloom filters—used to check whether an element or piece of data is present in a set
Index structures take up space in a database, so you need to be selective about what to index, and they do not take advantage of the underlying data distributions. I’ve worked in settings where an administrator or expert user carefully implements a strategy for building indexes for a data warehouse based on important and common queries.
Indexes are really models or mappings—for instance, a Bloom filter can be thought of as a classification problem. In a recent paper, Kraska and his collaborators approach indexing as a learning problem. As such, they are able to build indexes that take into account underlying data distributions, are smaller in size (thus allowing for a more liberal indexing strategy), and their indexes execute faster. Software and hardware for computation are getting cheaper and better, so using machine learning to create index structures is something that may indeed become routine.
This ties with a larger trend of using machine learning to improve software systems and even software development. In the future, we’ll have database administrators who have machine learning tools at their disposal, which would allow them to manage larger and more complex systems, and these ML tools will free them to focus on complex tasks that are harder to automate.
Here are some highlights from our conversation:
Why use machine learning to learn index structures
I think it used to be the case that if you know you have the key distribution, you could leverage that, but you need to build a very specialized system for that. Then, if the data distribution changes, you need to adjust the whole system. At the same time, any learning mechanism was in the past and way too expensive to do it.
Things have changed a little bit because compute is becoming much cheaper. Suddenly, using machine learning to train this mapping actually pays off. On one hand, the B-tree structures are composed of a whole bunch of “IF statements,” and in the past, multiplications were very expensive. Now multiplications are getting cheaper and cheaper. Scaling “IF statements” is hard, but scaling math operations is at least relatively easier. In essence, we can trade these “IF statements” for multiplications, and that’s actually why suddenly learning the data distribution pays off.
… For B-trees, for example, we saw speed-ups of up to roughly 2X. However, the indexes were up to two orders of magnitude smaller.
Why B-Trees are models. Image from Tim Kraska, used with permission. The future of data management systems
If this machine learning approach really works out, I think this might change the way database systems are built. … Maybe the database administrator (DBA) of the future becomes a machine learning expert.
… My hope is that the system can figure out what model to use, but maybe if you want to have the best performance and you know your data very well, I can see that maybe the DBA / machine learning expert chooses a certain type of model to tune the index.
… There was this Tweet, essentially saying that machine learning will change how we build core algorithms and data structures. I think this is currently still the better analogy.
Related resources:
- Tupleware—redefining modern data analytics: a Strata Data 2014 presentation by Tim Kraska
- Artificial intelligence in the software engineering workflow: a 2017 AI Conference keynote by Peter Norvig
- “A scalable time-series database that supports SQL“
- “Architecting and building end-to-end streaming applications“
- Data is only as valuable as the decisions it enables
In this episode of the Data Show, I spoke with Christine Hung, head of data solutions at Spotify. Prior to joining Spotify, she led data teams at the NY Times and at Apple (iTunes). Having led teams at three different companies, I wanted to hear her thoughts on digital transformation, and I wanted to know how she approaches the challenge of building, managing, and nurturing data teams.
I also wanted to learn more about what goes into building a recommender system for a popular consumer service like Spotify. Engagement should clearly be the most important metric, but there are other considerations, such as introducing users to new or “long tail” content.
Here are some highlights from our conversation:
Recommenders at Spotify
For us, engagement always comes first. At Spotify, we have a couple hundred people who are just focused on user engagement, and this is the group that creates personalized playlists, like Discover Weekly or your Daily Mix for you. We know our users love discovery and see Spotify as a very important platform for them to discover something new, but there are also times when people just want to have some music played in the background that fits the mood. But again, we don’t have a specific agenda in terms of what we should push for. We want to give you what you want so that you are happy, which is why we invested so much in understanding people through music. If we believe you might like some “long tail” content, we will recommend it to you because it makes you happy, but we can also do the same for the top 100 track if we believe you will enjoy them.
Music is like a mirror
Music is like a mirror, and it tells people a lot about who you are and what you care about, whether you like it or not. We love to say “you are what you stream,” and that is so true. As you can imagine, we invest a lot in our machine learning capabilities to predict people’s preference and context, and of course, all the data we use to train the model is anonymized. We take in large amounts of anonymized training data to develop these models, and we test them out with different uses cases, analyze results, and use the learning to improve those models.
Just to give you my personal example to illustrate how it works, you can learn a lot about me just by me telling you what I stream. You will see that I use my running playlist only during the weekend in early mornings, and I have a lot of children’s songs streamed at my house between 5 p.m. and 7 p.m. I also have a lot of tango and salsa playlists that I created and followed. So what does that tell you? It tells you that I am probably a weekend runner, which means I have some kind of affiliation for fitness; it tells you that I am probably a mother and play songs for my child after I get home from work; it also tells you that I somehow like tango and salsa, so I am probably a dancer, too. As you can see, we are investing a lot into understanding people’s context and preference so we can start capturing different moments of their lives. And, of course, the more we understand your context, your preference, and what you are looking for, the better we can customize your playlists for you.
Related resources:
- Music, the window into your soul: Christine Hung’s keynote at Strata Data NYC 2017
- “Transforming organizations through analytics centers of excellence”: Carme Artigas on helping enterprises transform themselves with big data tools and technologies.
- “A framework for building and evaluating data products”: Grace Huang on lessons learned in the course of machine learning product launches.
- “How companies can navigate the age of machine learning”: to become a “machine learning company,” you need tools and processes to overcome challenges in data, engineering, and models.
In this episode of the Data Show, I spoke with Neha Narkhede, co-founder and CTO of Confluent. As I noted in a recent post on “the age of machine learning,” data integration and data enrichment are non-trivial and ongoing challenges for most companies. Getting data ready for analytics—including machine learning—remains an area of focus for most companies. It turns out, “data lakes” have become staging grounds for data; more refinement usually needs to be done before data is ready for analytics. By making it easier to create and productionize data refinement pipelines on both batch and streaming data sources, analysts and data scientists can focus on analytics that can unlock value from data.
On the open source side, Apache Kafka continues to be a popular framework for data ingestion and integration. Narkhede was part of the team that created Kafka, and I wanted to get her thoughts on where this popular framework is headed.
Here are some highlights from our conversation:
The first engineering project that made use of Apache Kafka
If I remember correctly, we were putting Hadoop into a place at LinkedIn for the first time, and I was on the team that was responsible for that. The problem was that all our scripts were actually built for another data warehousing solution. The questions was, are we going to rewrite all of those scripts and now sort of make them Hadoop specific? And what happens when a third and a fourth and a fifth system is put into place?
So, the initial motivating use case was: ‘we are putting this Hadoop thing into place. That’s the new-age data warehousing solution. It needs access to the same data that is coming from all our applications. So, that is the thing we need to put into practice.’ This became Kafka’s very first use case at LinkedIn. From there, because that was very easy and I actually helped move one of the very first workloads to Kafka, it was hardly difficult to convince the rest of the LinkedIn engineering team to start moving over to Kafka.
So from there, Kafka adoption became pretty vital. Now, I think years down the line, all of LinkedIn runs on Kafka. It’s essentially the central nervous system for the whole company.
Microservices and Kafka
My own opinion of microservices is that it lets you add more money and turn it into software at a more constant rate by allowing engineers to focus on various parts of the application, by essentially decoupling a big monolith so that a lot of things can happen in parallel development of real applications.
… The upside is that it lets you move fast. It adds a certain amount of agility to an engineering organization. But it comes with its own set of challenges. And these were not very obvious back then. How are all these microservices deployed? How are they monitored? And, most importantly, how do they communicate with each other? The communication bit is where Kafka comes in. When you break a monolith, you break state. And you distribute that state across different machines that run all those different applications.
So now the problem is, ‘well, how do these microservices share that state? How do they talk to each other?’ Frequently, the expectation is that things happens in real time. The context of microservices where streams or Kafka comes in is in the communication model for those microservices. I should just say that there isn’t a one size fits all when it comes to communication patterns for microservices.
Related resources:
- Kafka: The Definitive Guide
- “Architecting and building end-to-end streaming applications“: Karthik Ramasamy on Heron, DistributedLog, and designing real-time applications.
- “Semi-supervised, unsupervised, and adaptive algorithms for large-scale time series“: Ira Cohen on developing machine learning tools for a broad range of real-time applications.
- “Building Apache Kafka from scratch“: Jay Kreps on data integration, event data, and the Internet of Things.
- I Logs
- “How companies can navigate the age of machine learning“: to become a “machine learning company,” you need tools and processes to overcome challenges in data, engineering, and models.
When I first discovered and started using Apache Spark, a majority of the use cases I used it for involved unstructured text. The absence of libraries meant rolling my own NLP utilities, and, in many cases, implementing a machine learning library (this was pre deep learning, and MLlib was much smaller). I’d always wondered why no one bothered to create an NLP library for Spark when many people were using Spark to process large amounts of text. The recent, early success of BigDL confirms that users like the option of having native libraries.
In this episode of the Data Show, I spoke with David Talby of Pacific.AI, a consulting company that specializes in data science, analytics, and big data. A couple of years ago I mentioned the need for an NLP library within Spark to Talby; he not only agreed, he rounded up collaborators to build such a library. They eventually carved out time to build the newly released Spark NLP library. Judging by the reception received by BigDL and the number of Spark users faced with large-scale text processing tasks, I suspect Spark NLP will be a standard tool among Spark users.
Talby and I also discussed his work helping companies build, deploy, and monitor machine learning models. Tools and best practices for model development and deployment are just beginning to emerge—I summarized some of them in a recent post, and, in this episode, I discussed these topics with a leading practitioner.
Here are some highlights from our conversation:
The state of NLP in Spark
Here are your two choices today. Either you want to leverage all of the performance and optimization that Spark gives you, which means you want to stay basically within the JVM, and you want to use a Java-based library. In which case, you have options that include OpenNLP, which is open source, or Stanford NLP, which requires licensing in order to use in a commercial product. These are older and more academically oriented libraries. So, they have limitations in performance and what they do.
Another option is to look at something like spaCy—a Python-based library that really has raised the bar in terms of usability, and the trade-offs between analytical accuracy and performance. But then your challenge is that you have your text in Spark, but to call the spaCy pipeline, you basically have to move the data from the JVM to a Python process, do some processing there, and send it back, which in practice means you take a huge performance hit because most of the processing you do is really moving strings between operating system processors.
… So, really what we were looking for is a solution to work on text directly, within a data frame. A tool that will take into account everything Spark gives in terms of caching, distributed computation, and the other optimizations. This enable users to basically run an NLP and machine learning pipeline directly on their text.
Enter Spark NLP Spark NLP. Image by David Talby, used with permission.
The core purpose of an NLP library is the ability to take text and then apply a set of annotations on the text. So, the basic annotations we ship in this initial version of Spark NLP include things like a tokenizer, a lemmatizer, sentence boundary detection, and paragraph boundary detection. Then on top of that, we include things like sentiment analysis, spell checker so we can auto-suggest corrections, and a dependency parser so we can not just know that we have a noun and a verb, but also that this verb talks about the specific noun, which is often semantically interesting. We also include named entity recognition algorithms.
Deploying and monitoring machine learning models in production
I think what’s happening is that people expect basic model development to be very similar to software development. When we started doing software development, we started it wrong. We assumed software engineering was a lot like civil engineering or mechanical engineering. It took a good 30 years until we said no, this is actually not the right model and we tried something different.
So, now I think we have the same paradigm error. People assume that, look, this is software. I build it, I deploy it on machine, I test it, I put it in production, therefore I should think of it like software, right? It’s something I build, I test it, it passes it, I deploy it, and it’s done, it moves to the ops team.
In reality, there really is a fundamental difference. What happens is once you put a machine learning model in production, at that point it starts degrading in terms of accuracy because of many, many things. Because the world changes, because the underlying population changes, the data you’re serving in production changes, because people react and basically try to game your model, because you have undocumented data dependencies on other systems. … I think you really need to be aware of the fact that when you deploy a machine learning model to production, in many cases really, you’re just starting your modeling process.
Related resources:
- Building pipelines for natural language understanding with Spark
- “Deep learning for Apache Spark”: Jason Dai on BigDL, a library for deep learning on existing data frameworks.
- When models go rogue: David Talby on hard-earned lessons about using machine learning in production
- “The current state of applied data science”
- “What are machine learning engineers?”: examining a new role focused on creating data products and making data science work in production.
In this episode of the Data Show, I spoke with Rhea Liu, analyst at China Tech Insights, a new research firm that is part of Tencent’s Online Media Group. If there’s one place where AI and machine learning are discussed even more than the San Francisco Bay Area, that would be China. Each time I go to China, there are new applications that weren’t widely available just the year before. This year, it was impossible to miss bike sharing, mobile payments seemed to be accepted everywhere, and people kept pointing out nascent applications of computer vision (facial recognition) to identity management and retail (unmanned stores).
I wanted to consult local market researchers to help make sense of some of the things I’ve been observing from afar. Liu and her colleagues have put out a series of interesting reports highlighting some of these important trends. They also have an annual report—Trends & Predictions for China’s Tech Industry in 2018—that Liu will discuss in her keynote and talk at Strata Data Singapore in December.
Here are some highlights from our conversation:
Machine learning and content distribution
Media consumption takes a large proportion of people’s everyday life here in China. Before, people learned their news from news portals and from editorial teams who served as the gatekeepers. People now trust machine learning algorithms with editorial and agenda setting. Apps like Toutiao have become very popular.
It’s been quite a surprise to most news portals and media professionals here in China. People are trying to find a balance between the traditional ways of content creation and the new ways of content distribution by aggregators fully powered by machines. Toutiao’s news recommendation engine is purely a black box to most people. … But users are spending more and more time on these types of platforms. And, machine-generated news feeds have become a big thing.
… So, it’s now becoming a content war again. After these algorithms improve the efficiency of content distribution, the battle may come down to what content you have.
Bike sharing
Bike sharing is kind of a new model adapted to Chinese society. … In between every subway station, there’s still several miles to go, where people still need to walk or maybe take a taxi. Bike sharing is being used to replace these other kinds of approaches.
There are two primary players. One is Mobike and the other one is Ofo, and they started with different models, actually. Ofo started a year or two earlier from a university campus. … It provided this kind of public bike rental system to users on campus. This was kind of the preliminary prototype of this model. Mobike started in a city.
These bike sharing companies have their GPS systems on the bikes, and the bikes have digital electronic locks that can be unlocked with an app on your phone. These technologies, combined together, can help them collect data as well as have a better management system of all the bikes they distribute over a city.
Smart cities
It’s still a maybe, but it’s very likely we are going to include things about smart cities in our 2018 reports. … This includes AR applications to help build better cities for urban planning. … Urban planning is a very complicated thing, and what we are missing there is, we can be a little bit left behind because of the lack of data. But now people have different types of data. For example, I know the ride sharing company Didi is collaborating with several city governments to help them do urban planning: by using data to better understand traffic, how to manage traffic light systems in the city, and also the bus system.
City governments at all levels are now collaborating with all these tech companies to explore applications of their data to improve the cities we have in China. … This is going to be a very important opportunity for the tech companies here in China, especially in terms of their data applications, and their contributions to society.
Related resources:
- “Vehicle-to-vehicle communication networks and smart cities”: Bruno Fernandez-Ruiz on the importance of building the ground control center of the future
- “How intelligent data platforms are powering smart cities”
- “Creating autonomous vehicle systems”: Shaoshan Liu on understanding AV technologies and how to integrate them
- Cars that coordinate with people: 2017 AI Conference keynote by Anca Dragan
In this episode of the Data Show, I spoke with Bruno Fernandez-Ruiz, co-founder and CTO of Nexar. We first met when he was leading Yahoo! technical teams charged with delivering a variety of large-scale, real-time data products. His new company is helping build out critical infrastructure for the emerging transportation sector.
While some question whether V2X communication is necessary to get to fully autonomous vehicles, Nexar is already paving the way by demonstrating how a vehicle-to-vehicle (V2V) communication network can be built efficiently. As Fernandez-Ruiz points out, there are many applications for such a V2V network (safety being the most obvious one). I’m particularly fascinated by what such a network, and the accompanying data, opens up for future, smarter cities. As I pointed out in a post on continuous learning, simulations are an important component of training AI applications. It seems reasonable to expect that the data sets collected by V2V networks will be useful for smart city planners of the future.
Here are some highlights from our conversation:
The many applications of a vehicle-to-vehicle network
Imagine if every vehicle on the road was equipped with a transponder that allowed it to connect to a network and say: ‘Hey, here I am. This is where I’m going. This is how fast I’ve gone. This is where I’ve been in the last 10 seconds. This is where I think I’m going to be in the next 10 seconds.’ Now imagine you were sharing that with all the vehicles around you so all these vehicles can predict, react, and even proact how they should behave on the road.
You can start solving for safety and for traffic congestion. You start solving for utilization of the road, for pollution, and many other problems. … That’s kind of the vision. I think with autonomous vehicles on the road, this will be even more important. You’ll have humans sharing the road with them—then, how does this mix of human and autonomous vehicles talk to each other?
… So, we’re trying to just build that vehicle-to-vehicle network. You can call it ‘ground traffic control.’ It’s similar to what happened in the air when radar and beaconing technology became available, and people said, ‘Well, we should probably connect to these things to be able to know where the planes are and tell them where to go.’
Redundancy and safety
There are many situations in which other autonomous technologies actually may fail, and I think that’s where vehicle-to-vehicle communication becomes both a necessary technology for the true future of complete automation. It’s also a redundant system for when that camera, or for when that radar, when whatever other sensors in the car fail.
Related resources:
- “Creating autonomous vehicle systems“: Understanding AV technologies and how to integrate them
- Cars that coordinate with people: 2017 AI Conference keynote by Anca Dragan
- “How big data and AI will reshape the automotive industry“
- “How intelligent data platforms are powering smart cities“
- “Why continuous learning is key to AI“: A look ahead at the tools and methods for learning from sparse feedback.
In this episode of the Data Show, I spoke with Carme Artigas, co-founder and CEO of Synergic Partners (a Telefonica company). As more companies adopt big data technologies and techniques, it’s useful to remember that the end goal is to extract information and insight. In fact, as with any collection of tools and technologies, the main challenge is identifying and prioritizing use cases.
As Artigas describes, one can categorize use cases for big data into the following types:
- Improve decision-making or operational efficiency
- Generate new or additional revenue
- Predict or prevent fraud (forecasting or minimizing risks)
Artigas has spent many years helping large organizations develop best practices for how to use data and analytics. We discussed some of the key challenges faced by organizations that wish to adopt big data technologies, centers of excellence for analytics, and AI in the enterprise.
Here are some highlights from our conversation:
Adopting big data analytics: Remaining key challenges
For me, the first challenge is that there’s a lack of skills across organizations. We know there’s a global shortage of analytic talent, so it’s not only challenging for a company to acquire the right talent, but also to make sure that this talent is accessible across the organization. It’s usually concentrated in some departments, and it’s very difficult to leverage those skills for the good of the entire organization.
The second challenge I see is lack of standards and lack of governance. You might find that every single data science team uses their own libraries or their own version of code or their own software tools. They are thinking about the benefit for a particular use case, and the best tools and the best models for that particular use case. But this compartmentalized approach cannot scale up; having a variety of versions of libraries and tools make it very, very difficult to industrialize and implement global solutions.
Finally, new skills you need to develop are not only on the technical side—they are mostly on the business side. Decision-makers need to make decisions in different ways. They need to make decisions based on data, based on facts.
Center of excellence for analytics
The analytics center of excellence is a team of business and technical people that can be internal, external, and even crowd sourced. They have some centralized capabilities and also some distributed capabilities and resources, creating a common (online) workspace where they share methodologies, tools, models, and techniques. The objective is to gain efficiency and be able to implement initiatives across to the different business units. We have two main components of these centers of excellence: the business transformation unit and the deployment units.
The business transformation unit (BTU) is the primary link of the center of excellence with the underlying business. We create ambassadors, and these ambassadors, who are part of the BTU, are responsible for identifying and prioritizing all business use cases. Then they connect with the deployment units—which we call cells—and these cells can grow organically during a project. So, first of all, the center of excellence must be connected with business. … We also create a centralized function called the ‘core team’ and an expansion unit called the ‘extended team.’ We have a few types of cells: the analytical cells, the operational cells, and the data utilization cells. So, it’s a way of concentrating the resources, gaining operational efficiency, having a center of know-how transferred to the rest of the organization, and ensuring best practices and methodologies.
… A center of excellence is not a physical place. The center of excellence is a network of people who can be distributed in different geographies.
Related resources:
- “The stages of enterprise IoT adoption“: Teresa Tung on building a business case for the Internet of Things
- “Data governance: What you need to know“
- “Turning big data into actionable insights“: Evangelos Simoudis on data mining, investing in data startups, and corporate innovation
- What is Artificial Intelligence?
In this episode of the Data Show, we look back to a recent conversation I had at the Spark Summit in San Francisco with Ion Stoica (UC Berkeley professor and executive chairman of Databricks) and Matei Zaharia (assistant professor at Stanford and chief technologist of Databricks). Stoica and Zaharia were core members of UC Berkeley’s AMPLab, which originated Apache Spark, Apache Mesos, and Alluxio.
We began our conversation by discussing recent academic research that would be of interest to the Apache Spark community (Stoica leads the RISE Lab at UC Berkeley, Zaharia is part of Stanford’s DAWN Project). The bulk of our conversation centered around machine learning. Like many in the audience, I was first attracted to Spark because it simultaneously allowed me to scale machine learning algorithms to large data sets while providing reasonable latency.
Here is a partial list of the items we discussed:
- The current state of machine learning in Spark.
- Given that a lot of innovation has taken place outside the Spark community (e.g., scikit-learn, TensorFlow, XGBoost), we discussed the role of Spark ML moving forward.
- The plan to make it easier to integrate advanced analytics libraries that aren’t “textbook machine learning,” like NLP, time series analysis, and graph analysis into Spark and Spark ML pipelines.
- Some upcoming projects from Berkeley and Stanford that target AI applications (including newer systems that provide lower latency, higher throughput).
- Recent Berkeley and Stanford projects that address two key bottlenecks in machine learning—lack of training data, and deploying and monitoring models in production.
[Full disclosure: I am an advisor to Databricks.]
Related resources:
- Spark: The Definitive Guide
- Advanced Analytics with Spark
- High-performance Spark
- Learning Path: Get Started with Natural Language Processing Using Python, Spark, and Scala
- Learning Path: Getting Up and Running with Apache Spark
- “The current state of applied data science”
In this episode of the Data Show, I spoke with Ken Stanley, founding member of Uber AI Labs and associate professor at the University of Central Florida. Stanley is an AI researcher and a leading pioneer in the field of neuroevolution—a method for evolving and learning neural networks through evolutionary algorithms. In a recent survey article, Stanley went through the history of neuroevolution and listed recent developments, including its applications to reinforcement learning problems.
Stanley is also the co-author of a book entitled Why Greatness Cannot Be Planned: The Myth of the Objective—a book I’ve been recommending to anyone interested in innovation, public policy, and management. Inspired by Stanley’s research in neuroevolution (into topics like novelty search and open endedness), the book is filled with examples of how notions first uncovered in the field of AI can be applied to many other disciplines and domains.
The book closes with a case study that hits closer to home—the current state of research in AI. One can think of machine learning and AI as a search for ever better algorithms and models. Stanley points out that gatekeepers (editors of research journals, conference organizers, and others) impose two objectives that researchers must meet before their work gets accepted or disseminated: (1) empirical: their work should beat incumbent methods on some benchmark task, and (2) theoretical: proposed new algorithms are better if they can be proven to have desirable properties. Stanley argues this means that interesting work (“stepping stones”) that fail to meet either of these criteria fall by the wayside, preventing other researchers from building on potentially interesting but incomplete ideas.
Here are some highlights from our conversation:
Neuroevolution today
In the state of the art today, the algorithms have the ability to evolve variable topologies or different architectures. There are pretty sophisticated algorithms for evolving the architecture of a neural network; in other words, what’s connected to what, not just what the weight of those connections are—which is what deep learning is usually concerned with.
There’s also an idea of how to encode very, very large patterns of connectivity. This is something that’s been developed independently in neuroevolution where there’s not a really analogous thing in deep learning right now. This is the idea that if you’re evolving something that’s really large, then you probably can’t afford to encode the whole thing in the DNA. In other words, if we have 100 trillion connections in our brains, our DNA does not have 100 trillion genes. In fact, it couldn’t have a 100 trillion genes. It just wouldn’t fit. That would be astronomically too high. So then, how is it that with a much, much smaller space of DNA, which is about 30,000 genes or so, three billion base pairs, how would you get enough information in there to encode something that’s 100 trillion parts?
This is the issue of encoding. We’ve become sophisticated at creating artificial encodings that are basically compressed in an analogous way, where you can have a relatively short string of information to describe a very large structure that comes out—in this case, a neural network. We’ve gotten good at doing encoding and we’ve gotten good at searching more intelligently through the space of possible neural networks. We originally thought what you need to do is just breed by choosing among the best. So, you say, ‘Well, there’s some task we’re trying to do and I’ll choose among the best to create the next generation.’
We’ve learned since then that that’s actually not always a good policy. Sometimes you want to explicitly choose for diversity. In fact, that can lead to better outcomes.
The myth of the objective
Our book does recognize that sometimes pursuing objectives is a rational thing to do. But I think the broader point that’s important here is there’s a class of discoveries for which it really is against your interest to frame what you’re doing in terms of an objective.
The reason we wrote the book is because … I started to realize this principle that ‘sometimes in order to make discovery possible, you have to stop having an objective’ speaks to people beyond just computer scientists who are developing algorithms. It’s an issue for our society and for institutions because there are many things we do that are driven by some kind of objective metric. It almost sounds like heresy to suggest that you shouldn’t do that.
It’s like an unquestioned assumption that exists throughout our culture that the primary route to progress is to set objectives and move toward those objectives and measure your performance with respect to those objectives. We began to think that given the results we have that are hard empirical results, that it is important to counterweight this belief that pervades society with a counter argument that points out that there are cases where this is actually a really bad idea.
The thing I learned more and more talking to different groups is that this discussion is not being had. We’re not talking about this, and I think it’s a very important discussion because our institutions are geared away from innovation because they are so objectively driven. We could do more to foster innovation if we recognize this principle. A lot of people want this security blanket of an objective because they don’t trust anything that isn’t driven by an objective.
Actually, it turns out there are principled ways of exploring the world without an objective. In other words, it’s not just random and the book is about that. It’s about how smart ways of exploring in a non-objective way can lead to really, really important results. We just wanted to open up that conversation ‘society wide’ and not just have it narrowly within the field of computer science because it is such an important conversation to have.
Related resources:
- “Neuroevolution: A different kind of deep learning“: a must-read overview by Ken Stanley
- “Why continuous learning is key to AI“: A look ahead at the tools and methods for learning from sparse feedback.
- Ray: A distributed execution framework for emerging AI applications: 2017 Strata Data keynote by Michael Jordan
- Deep reinforcement learning for robotics: 2016 AI Conference presentation by Pieter Abbeel
- Cars that coordinate with people: 2017 AI Conference keynote by Anca Dragan
In this episode of the Data Show, I spoke with Robert Nishihara and Philipp Moritz, graduate students at UC Berkeley and members of RISE Lab. I wanted to get an update on Ray, an open source distributed execution framework that makes it easy for machine learning engineers and data scientists to scale reinforcement learning and other related continuous learning algorithms. Many AI applications involve an agent (for example a robot or a self-driving car) interacting with an environment. In such a scenario, an agent will need to continuously learn the right course of action to take for a specific state of the environment.
What do you need in order to build large-scale continuous learning applications? You need a framework with low-latency response times, one that is able to run massive numbers of simulations quickly (agents need to be able explore states within an environment), and supports heterogeneous computation graphs. Ray is a new execution framework written in C++ that contains these key ingredients. In addition, Ray is accessible via Python (and Jupyter Notebooks), and comes with many of the standard reinforcement learning and related continuous learning algorithms that users can easily call.
As Nishihara and Moritz point out, frameworks like Ray are also useful for common applications such as dialog systems, text mining, and machine translation. Here are some highlights from our conversation:
Tools for reinforcement learning
Ray is something we’ve been building that’s motivated by our own research in machine learning and reinforcement learning. If you look at what researchers who are interested in reinforcement learning are doing, they’re largely ignoring the existing systems out there and building their own custom frameworks or custom systems for every new application that they work on.
… For reinforcement learning, you need to be able to share data very efficiently, without copying it between multiple processes on the same machine, you need to be able to avoid expensive serialization and deserialization, and you need to be able to create a task and get the result back in milliseconds instead of hundreds of milliseconds. So, there are a lot of little details that come up.
… In fact, people often use MPI along with lower-level multi-processing libraries to build the communication infrastructure for their reinforcement learning applications.
Scaling machine learning in dynamic environments
I think right now when we think of machine learning, we often think of supervised learning. But a lot of machine learning applications are changing from making just one prediction to making sequences of decisions and taking sequences of actions in dynamic environments.
The thing that’s special about reinforcement learning is it’s not just the different algorithms that are being used, but rather the different problem domain that it’s being applied to: interactive, dynamic, real-time settings bring up a lot of new challenges.
… The set of algorithms actually goes even a little bit further. Some of these techniques are even useful in, for example, things like text summarization and translation. You can use these techniques that have been developed in the context of reinforcement learning to better tackle some of these more classical problems [where you have some objective function that may not be easily differentiable].
… Some of the classic applications that we have in mind when we think about reinforcement learning are things like dialogue systems, where the agent is one participant in the conversation. Or robotic control, where the agent is the robot itself and it’s trying to learn how to control its motion.
… For example, we implemented the evolution algorithm described in a recent OpenAI paper in Ray. It was very easy to port to Ray, and writing it only took a couple of hours. Then we had a distributed implementation that scaled very well and we ran it on up to 15 nodes.
Related resources:
- Why continuous learning is key to AI: A look ahead at the tools and methods for learning from sparse feedback
- Ray: A distributed execution framework for emerging AI applications: A Strata Data keynote by Michael Jordan
- Deep reinforcement learning for robotics: A 2016 Artificial Intelligence Conference presentation by Pieter Abbeel
- Cars that coordinate with people: A 2017 Artificial Intelligence Conference keynote by Anca Dragan)
- Introduction to reinforcement learning and OpenAI Gym
- Reinforcement learning explained
In this episode of the Data Show, I spoke with Soumith Chintala, AI research engineer at Facebook. Among his many research projects, Chintala was part of the team behind DCGAN (Deep Convolutional Generative Adversarial Networks), a widely cited paper that introduced a set of neural network architectures for unsupervised learning. Our conversation centered around PyTorch, the successor to the popular Torch scientific computing framework. PyTorch is a relatively new deep learning framework that is fast becoming popular among researchers. Like Chainer, PyTorch supports dynamic computation graphs, a feature that makes it attractive to researchers and engineers who work with text and time-series.
Here are some highlights from our conversation:
The origins of PyTorch
TensorFlow addressed one part of the problem, which is quality control and packaging. It offered a Theano style programming model, so it was a very low-level deep learning framework. … There are a multitude of front ends that are trying to cope with the fact that TensorFlow is a very low-level framework—there’s TF-slim, there’s Keras. I think there’s like 10 or 15, and just from Google there’s probably like four or five of those.
On the Torch side, the philosophy has always been slightly different than Theano. I see TensorFlow as a much better Theano-style framework, and on the Torch side we had a philosophy that we want to be imperative, which means that you run your computation immediately. Debugging should be butter smooth. The user should never have trouble debugging their programs, whether they use a Python debugger or something like the GDB or something else.
… Chainer was a huge inspiration. PyTorch is inspired primarily by three frameworks. Within the Torch community, certain researchers from Twitter built an auxiliary package called Autograd, and this was actually based on a package called Autograd in the Python community. Like Chainer, Autograd and Torch Autograd, all used a certain technique called tape-based automatic differentiation: that is, you have a tape recorder that records what operations you have performed and then it replays it backward to compute your gradients. This is a technique that is not used by any of the other major frameworks except PyTorch and Chainer. All of the other frameworks use what we call a static graph—that is, the user builds a graph, then they give that graph to an execution engine that is provided by the framework, and the framework executes it. It can analyze it ahead of time.
These are very two different techniques. The tape-based differentiation gives you easier debuggability, and it gives you certain things that are more powerful (e.g., dynamic neural networks). The static graph-based approach gives you easier deployment to mobile, easier deployment to more exotic architectures, the ability to do compiler techniques ahead of time, and so on.
Deep learning frameworks within Facebook
Internally at Facebook, we have a unified strategy. We say PyTorch is used for all of research and Caffe 2 is used for all of production. This makes it easier for us to separate out which team does what and which tools do what. What we are seeing is, users first create a PyTorch model. When they are ready to deploy their model into production, they just convert it into a Caffe 2 model, then ship into either mobile or another platform.
PyTorch user profiles
PyTorch has gotten its biggest adoption from researchers, and it’s gotten about a moderate response from data scientists. As we expected, we did not get any adoption from product builders because PyTorch models are not easy to ship into mobile, for example. We also have people who we did not expect to come on board, like folks from OpenAI and several universities.
Related resources:
- Building intelligent applications with deep learning and TensorFlow
- BigDL: Deep learning for Apache Spark
- MXNet: Deep learning that’s easy to implement and easy to scale
- PyTorch: A flexible and intuitive framework for deep learning
- Scaling Machine Learning
In this episode of the Data Show, I spoke with Evangelos Simoudis, co-founder of Synapse Partners and a frequent contributor to O’Reilly. He recently published a book entitled The Big Data Opportunity in Our Driverless Future, and I wanted get his thoughts on the transportation industry and the role of big data and analytics in its future. Simoudis is an entrepreneur, and he also advises and invests in many technology startups. He became interested in the automotive industry long before the current wave of autonomous vehicle startups was in the planning stages.
Here are some highlights from our conversation:
Understanding the automotive industry
The more I started spending time with the automotive industry, the more I came to realize that, because of the autonomous vehicle technology and because of various forms of mobility services, which are stemming from new business models, the incumbent automotive industry is in significant risk of being disrupted.
If you were to look at the automotive industry, the first thing that is very striking is that there’s a small number of very large companies that control a number of different labels. With GM, we talk about Chevy, we talk about Buick, we talk about Opel in Europe. There are a very small number of companies that control this trillion dollar industry.
The other thing that is interesting is that these companies are responsible for designing the vehicle, manufacturing it, assembling it, post-manufacturing, and then creating demand, whereas the sale of the vehicle is done through the dealers. And they’re paying relatively little attention to what happens post-sale. So, that means there is a relatively little understanding of consumer behavior.
The third observation is that the reason there are so few of these companies is because starting one is very capital intensive. And if you look at how much money, for example, a company like Tesla has been able to raise, you get a sense of what kind of capital is necessary. And the next point is that even though there is a lot of capital that’s being raised, in the end this is a relatively low margin business. Where you try to make it up is in volume. That’s why, if you look at all these corporations, they have extremely sophisticated supply chains, extremely sophisticated manufacturing lines, highly optimized, because they are working on maintaining these margins.
Infrastructure for autonomous vehicles
A vehicle needs to know very much what’s happening around it. So that means it needs to receive signals from roads, bridges, other vehicles. … The term people use is V2X or vehicle-to-everything communication.
It will take a very long time to have the preponderance of vehicles being autonomous. So, we need infrastructure that will enable cars to safely operate in a hybrid world between autonomous vehicles and manually operated vehicles. I think the experiments that today involve just a few tens of cars will expand over the next few years. And I think the result of those experiments will give us an understanding and appreciation of the investments that we need to make and how to prioritize them, as well as the regulations that we will need to institute in order to have this type of hybrid environment operate safely.
AI and big data
The argument that I’m making, and this actually comes from my education on AI and my work on AI since the mid ‘80s, is that while machine learning is important, I think everybody needs to appreciate that it’s not only about machine learning. In order to bring to realization an autonomous vehicle, you need more than machine learning. And, of course, within machine learning we have neural network learning and particularly deep learning, and these are very important areas.
But people need to realize that an autonomous vehicle requires the ability to plan, requires the ability to reason, to represent knowledge, to search. All of these are components of AI. What I’m hoping to impart is that it’s not only about machine learning and particularly not about deep learning. The popular press, I think, is leading everybody to believe that it’s only about deep learning.
There is the autonomous driving technology and then the data cloud, where big data gets processed, stored, and analyzed. I think we will have multiple cloud providers. In fact, I’m betting on that through my investments in the space. I think that those cloud providers will be in the application layer. So, those cloud providers may be utilizing infrastructures from the likes of Microsoft or Amazon or other generic clouds.
Related resources:
- Creating autonomous vehicle systems: Shaoshan Liu on the technology behind self-driving cars
- Big data opportunities in next-generation mobility: Evangelos Simoudis’ presentation at Strata Data San Jose
- Turning big data into actionable insights: Evangelos Simoudis on a previous episode of the Data Show
- A new value chain for next-generation mobility
- Exploring the process of insight generation
