Austrian Ai Podcast: Recent Episodes

Manuel Pasieka

Guest Interviews, discussing the possibilities and potential of AI in Austria.

Question or Suggestions, write to austrianaipodcast@pm.me

Hello and welcome back to the AAIP

If you are an active Machine Learning engineer or are simply interested in Large Language models, I am sure you have seen the discussions around quantized models and all kind of new frameworks that have appeared recently and achieve astonishing inference performance of LLM's on consumer devices.

If you are curious how modern Large Language Models with their billions of parameters can run on a simple laptop or even an embedded device, than this episode is for you.

Today I am talking to Eldar Kurtic, researcher in the Alistarh group at the IST in lower Austrian and senior research engineer at the American startup Neural Magic.

Eldar's research focuses on optimizing Inference of Deep Neural Networks. On the show he is going to explain in depth show sparsity and quantization works, and how they can be applied to accelerate inference of big models, like LLM's on devices with limited resources.

Because of the length of the interview, I decided to split it into two parts.

This one, the first part, is going to focus on sparsity to reduce model size and enable faster inference by reducing the amount of memory and compute that is needed to store and run models.

The second part is going to focus on quantization as a mean to find representations of models with lower numeric precision that require less memory to store and process, while retaining accuracy.

In this first part about sparsity, Eldar will explain fundamental concepts like structured and unstructured sparsity. How and why they work and how currently we can achieve performant inference of unstructured sparsity only on CPU's and far less on GPU's.

We will discuss how to achieve crazy numbers of up to 95% unstructured sparsity while retaining the accuracy of models, but why it is difficult to leverage this under quoutes, reduction in model size, to actually accelerate model inference.

Enjoy.

AAIP Community

Join our discord server and ask guest directly or discuss related topics with the community.

https://discord.gg/5Pj446VKNU

References

Eldar Kurtic: https://www.linkedin.com/in/eldar-kurti%C4%87-77963b160/

Neural Magic: https://neuralmagic.com/

IST Austria Alistarh Group: https://ist.ac.at/en/research/alistarh-group/

Summary

Today on the show I am talking to Veronika Vishnevskaia. Solution Architect at ONTEC where she specialises in building RAG based Question-Answering systems.

Veronika will provide a deep dive into all relevant steps to build a Question-Answering system. Starting from data extraction and transformation, followed by text embedding, chunking and hybrid retrieval to strategies and last but not least methods to mitigate hallucinations of LLMs during the answer creation.

AAIP Community

Join our discord server and ask guest directly or discuss related topics with the community.

https://discord.gg/5Pj446VKNU

TOC

00:00:00 Beginning

00:03:33 Guest Introduction

00:08:51 Building Q/A Systems for businesses

00:16:27 RAG: Data extraction & pre-processing

00:26:08 RAG: Chunking & Embedding

00:36:13 RAG: Information Retrieval

00:48:59 Hallucinations

01:02:21 Future RAG systems

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/
Belichberg GmbH: Software that Saves the Planet: The Future of Energy Begins Here - https://belichberg.com/

References

Veronika Vishnevskaia - https://www.linkedin.com/in/veronika-vishnevskaia/

Ontec - www.ontec.at

Review Hallucination Mitigation Techniques: https://arxiv.org/pdf/2401.01313.pdf

Aleph-Alpha: https://aleph-alpha.com/de/technologie/

Summary

Today on the show I am talking to Manuel Reinsperger, Cybersecurity Expert and Penetration Tester. Manuel will provide us an introduction into the topic of Machine Learning Security with an emphasis on Chatbot and Large Language Model security.

We are going to discuss topics like AI Red Teaming that focuses on identifying and testing AI systems within an holistic approach for system security. Another major theme of the episode are different Attack Scenarios against Chatbots and Agent systems.

Manuel will explain to use, what Jailsbreak are and methods to exfiltrate information and cause harm through direct and indirect prompt injection.

Machine Learning security is a topic I am specially interested in and I hope you are going to enjoy this episode and find it useful.

AAIP Community

Join our discord server and ask guest directly or discuss related topics with the community.

https://discord.gg/5Pj446VKNU

TOC

00:00:00 Beginning

00:02:05 Guest Introduction

00:05:16 What is ML Security and how does it differ from Cybersecurity?

00:25:56 Attacking chatbot systems

00:41:12 Attacking RAGs with Indirect prompt injection

00:54:43 Outlook on LLM security

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/
Belichberg GmbH: Software that Saves the Planet: The Future of Energy Begins Here - https://belichberg.com/

References

Manuel Reinsperger - https://manuel.reinsperger.org/

Test your prompt hacking skills: https://gandalf.lakera.ai/

Hacking Bing Chat: https://betterprogramming.pub/the-dark-side-of-llms-we-need-to-rethinInjectGPT: k-large-language-models-now-6212aca0581a

AI-Attack Surface: https://danielmiessler.com/blog/the-ai-attack-surface-map-v1-0/

https://blog.luitjes.it/posts/injectgpt-most-polite-exploit-ever/

https://github.com/jiep/offensive-ai-compilation

AI Security Reference List: https://github.com/DeepSpaceHarbor/Awesome-AI-Security

Prompt Injection into GPT: https://kai-greshake.de/posts/puzzle-22745/

Summary

At the end of last year, the EU-AI Act was finalized and it spawned many discussions and a lot of doubts about the future of European AI companies.

Today on the show Peter Jeitschko, founder of JetHire an AI based recruiting platform that uses Large Language models to help recruiters find and work with candidates, talks about this perspective on the AI-Act.

We talk about the impact of the EU AI-Act on their platform, and how it falls into a high-risk use-case under the new regulation. Peter describes how the AI-Act forced them to create their company in the US and what he believes are the downsides of the EU regulation.

He describes his experience, that the EU regulations hinder innovation in Austria and Europe and how it increases legal costs and uncertainty, resulting in decision makers shying away in building and applying modern AI systems.

I think this episode is valuable for decision makers and founders of AI companies, that are affected by the upcoming AI Act and struggle to make sense of it.

AAIP Community

Join our discord server and ask guest directly or discuss related topics with the community.

https://discord.gg/5Pj446VKNU

TOC

00:00:00 Beginning

00:03:09 Guest Introduction

00:04:45 A founders perspective on the AI Act

00:13:45 JetHire - A recruiting platform affected the the AI Act

00:19:58 Achieving regulatory goals with good engineering

00:35:22 The mismatch between regulations and real world applications

00:48:12 European regulations vs. global AI services

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/
Belichberg GmbH: Software that Saves the Planet: The Future of Energy Begins Here - https://belichberg.com/

References

Peter Jeitschko - https://www.linkedin.com/in/pjeitschko/

Peter Jeitschko - https://peterjeitschko.com/

JetHire - https://jethire.ai/

https://www.holisticai.com/blog/requirements-for-high-risk-ai-applications-overview-of-regulations

Summary

For the last two years AI has been flooded with news about LLMs and their successes, but how many companies are actually making use of them in their products and services?

Today on the show I am talking to Markus Keiblinger, Managing partner of Texterous. A startup that focus on building custom LLM Solutions to help companies automate their business.

Markus will tell us about his experience when talking and working with companies building such LLM focused solutions.

Telling us about the expectations companies have on the capabilities of LLMs, as well on what companies need to have in order to be successfully implementing LLM projects.

We will discuss how Textorous has successfully focused on Retriever Augmented Generation (RAG) use cases.

RAGs is a mechanism that makes it possible to provide information to an LLM in a controlled menner, so the LLM can answer questions or follow instructions making use of that information. This enables companies to make use of their data to solve problems with LLMs, without having to train or even fine-tune models. On the show, Markus will tell us of one of these RAG projects and we will contrast building a RAG system based on Service Provider offerings like OpenAI or self hosted open source alternatives.

Last but not least, we talk about new use cases emerging with multi-modal Models, and the long term perspective that exists for custom LLM Solutions Providers like them in focusing on building integrated solutions.

AAIP Community

Join our discord server and ask guest directly or discuss related topics with the community.

https://discord.gg/5Pj446VKNU

TOC

00:00:00 Beginning

00:03:31 Guest Introduction

00:06:40 Challenges of applying AI in medical applications

00:17:56 Homogeneous Ensemble Methods

00:25:50 Combining base model predictions

00:40:14 Composing Ensembles

00:52:24 Explainability of Ensemble Methods

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/
Belichberg GmbH: Software that Saves the Planet: The Future of Energy Begins Here - https://belichberg.com/

References

Markus Keiblinger: https://www.linkedin.com/in/markus-keiblinger
Texterous: https://texterous.com
Book: Conversations Plato Never Captured - but an AI did: https://www.amazon.de/Conversations-Plato-Never-Captured-but/dp/B0BPVS9H9R/

Summary

Hello and welcome back to the Austrian Artificial Intelligence Podcast in 2024.

With this episode we start into the third year of the podcast. I am very happy to see that the number of listeners has been growing steadily since the beginning and I want to thank you dear listeners for coming back to the podcast and sharing it with your friends.

Gabriel is a Bioinformatician at the Austrian Institute of Technology and is going to explain his work on ensemble methods and their application in the medical domain.

For those not familiar with the term, an Ensemble is a combination of individual base models that are combined with the goal to outperform each individual model.

So the basic idea is, that one combines multiple models that each have their strength and weaknesses into a single ensemble that in the best case has all the strengths without the weaknesses.

We have seen one type of ensemble methods in the past. These where homogeneous ensemble methods like federated learning, where one trains the same algorithm multiple times by multiple parties or different subsets of the data, for performance reasons or in order to combine model weights without sharing the training data.

Today, Gabriel will talk about heterogeneous ensembles that are a combination of different models types and their usage in medical applications. He will explain how one can use them to increase the robustness and the accuracy of predictions. We will discuss how to select and create compositions of models, as well how to combine the different predictions of the individual base models in smart ways that improve their accuracy over simply methods like averaging over majority voting.

AAIP Community

Join our discord server and ask guest directly or discuss related topics with the community.

https://discord.gg/5Pj446VKNU

TOC

00:00:00 Beginning

00:03:31 Guest Introduction

00:06:40 Challenges of applying AI in medical applications

00:17:56 Homogeneous Ensemble Methods

00:25:50 Combining base model predictions

00:40:14 Composing Ensembles

00:45:57 Explainability of Ensemble Methods

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/
Belichberg GmbH: Software that Saves the Planet: The Future of Energy Begins Here - https://belichberg.com/

References

Gabriel Alexander Vignolle - https://www.linkedin.com/in/gabriel-alexander-vignolle-385b141b6/

Publications - https://publications.ait.ac.at/en/persons/gabriel.vignolle

Molecular Diagnostics - https://molecular-diagnostics.ait.ac.at/

Summary

In our bodies, the Immune system is detecting foreign pathogens or cancer cells, called antigens, with the help of antibody proteins that detect and physically attach to the surface of those cells.

Unfortunately our immune system is not perfect and does not detect all antigens, meaning that the immune system does not have all antigens it would need to detect all cancer cells for example.

Modern cancer therapies like CAR T-cells therapy therefor introduces additional antibody proteins into the system. This is still not enough to beat cancer, because cancer is a very diverse decease with a high variation of mutations between patients, and the antibodies used in CAR T-cell therapy are developed to be for a cancer type or patient group, but not for individual patience.

Today on the austrian AI podcast I am talking to Moritz Schäfer who is working on applying Diffusion Models to predict protein structures that support the development of patient specific, and therefore cancer mutation specific antibodies. This type of precision medicine would enable a higher specificity of cancer Therapie and will hopefully improve Treatment outcome.

Existing DL systems like Alpha Fold and alike fall short in predicting the structure of antibody binding sites, primarily due to lack of training data. So there room for improvement, and Moritz work is focused on applying Diffusion Models (so models like DALL-E or Stable Diffusion), which are most well known for their success in generating images, to problem of protein prediction. Diffusion models are generative models that generate samples from their training distribution based on an iterative process of several hundred steps. Where one starts, in case of image generation from pure noise, and in each step replaces noise with something that is closer to the training data distribution.

In Moritz work, they apply classifier guided Diffusion models to generate 3d antibody protein structures.

This means that in the iterative process of a diffusion model where in each step small adjustments are performed, a classifier nudges the changes towards increasing the affinity of the predicted protein to the specific antigen.

TOC

00:00:00 Beginning

00:03:23 Guest Introduction

00:06:37 The AI Institute at the UniWien

00:07:57 Protein Structure Prediction

00:10:57 Protein Antibodies in Caner Therapy

00:16:17 How precision medicine is applied in cancer Therapy

00:22:17 Lack of training data for antibody protein design

00:30:44 How Diffusion models can be applied in protein design

00:46:06 Classifier based Diffusion Models

00:51:18 Future in prediction medicine

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

Belichberg GmbH: We do digital transformations as your innovation partner - https://belichberg.com/

References

Moritz Schaefer - https://www.linkedin.com/in/moritzschaefer/

Unser Institut - https://www.meduniwien.ac.at/ai/de/contact.php

Lab website - https://www.bocklab.org/

LLM bio paper: https://www.biorxiv.org/content/10.1101/2023.06.16.545235v1

Diffusion Models - https://arxiv.org/pdf/2105.05233.pdf

Diffusion Models (Computerphile) - https://www.youtube.com/watch?v=1CIpzeNxIhU

Summary

Today on the show I am talking to Martin Huber Co-Founder and CEO of AMRAX.

We will talk about their product Metaroom; an AI application that is build on-top of consumer smartphones and makes it possible to create a digital twins of buildings for indoor user cases, like interior and light design.

We will focus less on algorithms and Machine Learning Methods, but on the impact that sensors and hardware platform have on the AI applications that can be build on top of them.

Martin will explain how Apple's LiDAR Sensors, available in their pro devices, in combination with the Apple Roomplan API are a unique and powerful platform to build AI applications, but at the same time forces one to focus on vertical integration and solutions. We will discuss how as an AI startup in this space one has to be super focused to be successful.

TOC

00:00:00 Beginning

00:02:57 Guest Introduction

00:05:08 Building hardware vs. writing software

00:07:12 AMRAX & Metaroom by AMRAX

00:13:51 3D Reconstruction with and without LiDAR

00:24:45 Data processing on device & in the cloud

00:30:55 Strategic positioning as a startup

00:37:11 Digital twin for smart home use cases

00:43:47 Future LiDAR sensors and their impact

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

Belichberg GmbH: We do digital transformations as your innovation partner - https://belichberg.com/

References

Martin Huber - https://www.linkedin.com/in/martin-huber-b940a084/

Lidar - https://amrax.ai/news/power-of-lidar

Metaroom - https://www.linkedin.com/showcase/metaroom-by-amrax/

Apples Roomplan - https://developer.apple.com/augmented-reality/roomplan/

Sony LiDAR Sensors - https://www.sony-semicon.com/en/news/2023/2023030601.html

Summary

I am sure that most of you are familiar with the training paradigm of supervised and unsupervised learning. Where in the case of supervised learning one has a label for each training datapoint and in the unsupervised situation there are no labels.

Although there can be exceptions, everyone is well advise to perform supervised training when ever possible. But where to get those labels for your training data if traditional labeling strategies, like manual annotations are not possible?

Well often you might not have perfect labels for your data, but you have some idea what those labels might be.

And this, my dear listener is exactly the are of weak supervision.

Today on the show I am talking to Andreas Stephan who is doing is PhD in Natural Language Processing at the University of Vienna in the Digital Text Sciences group led by Professor Benjamin Roth.

Andreas will explain about his recent research in the area of weak supervision as well how Large Language Models can be used as weak supervision sources for image classification tasks.

TOC

00:00:00 Beginning

00:01:38 Weak supervision a short introduction (by me)

00:04:17 Guest Introduction

00:08:48 What is weak supervision?

00:16:02 Paper: SepLL: Separating Latent Class Labels from Weak Supervision Noise

00:26:28 Benefits of priors to guide model training

00:29:38 Data quality & Data Quantity in training foundation models

00:36:10 Using LLM's for weak supervision

00:46:51 Future of weak supervision research

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/
Belichberg GmbH: We do digital transformations as your innovation partner - https://belichberg.com/

References

Andreas Stephan - https://andst.github.io/
Stephan et al. "SepLL: Separating Latent Class Labels from Weak Supervision Noise" (2022) - https://arxiv.org/pdf/2210.13898.pdf
Gunasekar et al. "Textbooks are all you need" (2023) - https://arxiv.org/abs/2306.11644
Introduction into weak supervision: https://dawn.cs.stanford.edu/2017/07/16/weak-supervision/

Summary

Personalized Medicine has the goal to improve the efficiency of medical treatments that is particularly import for deceases like cancer.

The basic premise of personalized medicine is that by understanding what distinguishes different patients from one another, we can design patience specific treatment plans that are more effective then giving the same treatment to everyone.

These patience differences can be studied from different perspectives, like a patience gender, their age, life style or more modern state of the art modalities, like a patience genome and epigenetic information.

Today on the show I am talking to Daria Romanovskaia an PhD Candidate in the Bocklab at the Research Center for Molecular Medicine in Vienna.

Daria describes how she applied machine learning methods on genetic and epigenetic data to identify patterns in gene regulation that can be used to create bio-makers that enable a targeted approach to decease treatment.

We will discuss how representation learning is used to perform dimensionality reduction and how clustering in combination with biological understanding of deceases can be used to identify relevant up and down regulated genes.

TOC

00:00:00 Beginning

00:00:27 Episode Introduction

00:01:47 Guest Introduction

00:03:57 What is precision medicine?

00:06:35 Short introduction into genetics

00:13:25 High dimensional data and how to work around it

00:17:16 Variational Autoencoders

00:24:19 Topic Modelling

00:32:19 Analyzing embedding clusters

00:35:56 Combining data analysis and domain expertise

00:39:32 The human cell atlas

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

Belichberg GmbH: We do digital transformations as your innovation partner - https://belichberg.com/

References

https://www.linkedin.com/in/daria-romanovskaia-b37b2a8a/ - Daria Romanovskaia

https://www.bocklab.org/ - Bock research lab at CeMM

https://www.humancellatlas.org/ - Human Cell Atlas

https://cemm.at/research/groups/christoph-bock-group

Summary

Did you ever had the experience that you where training a network, investing a lot of time in finding the right hyper parameters and testing different initializations to push that validation accuracy over certain threshold? Only to then find out when putting the model into production, that it significantly underperforms?

If you did, then you experienced one common problem with deep neural networks. The performance gap between in and out-of distribution generalization.

Today on the show PhD Rahim Entezari is giving us a wonderful tour through his PhD journey investigating ways to understand and improve generalization performance of deep neural networks.

Rahim will explain how one can improve generalization by different methods in data or in parameter space.

We will discuss how using different forms of sparsity, or the efficient creation of deep ensemble networks by permutation of network configurations can improve generalization from a parameter space perspective.

Or, from a data perspective where we discuss how data quality and data diversity effects the generalization performance of modern deep neural networks.

I hope you enjoy this interview, full of interesting concepts and ideas from deep learning theory.

TOC

00:00:00 Introduction

00:02:18 Background Knowledge

00:06:56 Guest Introduction

00:12:35 Generalization from a Data or Parameter Perspective

00:16:21 In and out of distribution Generalization

00:20:30 Structured and Unstructured Sparsity

00:29:55 Generalization in Parameter space

00:46:56 Generalization in Data space

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

Belichberg GmbH: We do digital transformations as your innovation partner - https://belichberg.com/

References

Rahim Entezari - https://www.linkedin.com/in/rahimentezari/

Episode

Today I am trying a new podcast format. This episode is a collection of short interviews taken live at the DSC DACH 2023 conference that has taken place in Vienna from the 20 to 21 of April.

The DSC DACH conference is a technology conference had a diverse set of talks and tracks, from industrial applications of AI to the next generation of deep learning methods. The motto of this years iteration was "Save the world through data". An ambitious goal, but I think some of the speakers had interesting ideas and I had a good time listening to their talks.

The goal of this new episode format my dear listener, is to give you a chance to get a glimpse of the type of talks and content that is presented at the event. So you know what you missed, can look into the things that interest you further or even join next years even.

I am curious what you think about this episode format, as I wan to repeat in the future to provide a type of live coverage of AI events, in addition to the normal guest interviews. Drop me an email or write me on LinkedIn if you have a strong opinion on the topic.

But thats enough for the intro and lets begin with the actual interviews.

I start with a short Interview with Aleksandar Djordjevic, on of the organizers of this conference who is going to tell us about its origin and what they want to achieve with the conference going forward.

Next I am speaking to William Amminger from Lumos Student Data Consulting who where presenting a project they are working on, and William was so kind to take the time, to answer a few questions about Lumos.

Next in line is Alexandra Ebert from Mostly.AI about open synthetic data and what role it plays in with respect to privacy and fairness.

Next is Stefan Papp about ways AI can help to battle climate change and the role Data Scientists and Tech Entrepreneurs play in finding ways to reduce a climate disaster.

Next I am talking to Danko Nikolic about a talk that he gave on the next generation of AI systems, that go beyond deep learning.

And last but not least, I am talking to Nicole Weinert about her work, in applying machine learning to empower education and therapeutic applications.

TOC

00:00:00 Introduction

00:01:33 Aleksandar Djordjevic on DSC DACH 2023

00:17:11 William Amminger on Lumos Student Data Consulting

00:24:58 Alexandra Ebert on Data Privacy and Fairness

00:33:08 Stefan Papp on Using Data Science to prevent a climate disaster

00:39:19 Danko Nikolic on next generation AI Systems beyond eep learning

00:47:07 Nicole Weinert on ML for education and therapeutic applications

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

Belichberg GmbH: We do digital transformations as your innovation partner - https://belichberg.com/

References

DSCDACH 2023 - https://dscdach.com/index.html
Aleksandar Linc-Djordjevic - https://www.linkedin.com/in/aleksandar-linc-djordjevic/
William Amminger - https://www.linkedin.com/in/william-amminger-619a8517a/
Lumos Student Data Consulting - https://www.linkedin.com/company/lumos-student-data-consulting/
Alexandra Ebert - https://www.linkedin.com/in/alexandraebert/
Stefan Papp - https://www.linkedin.com/in/stefanpapp/
Danko Nikolic - https://www.linkedin.com/in/danko-nikolic/
Nicole Weinert - https://www.linkedin.com/in/nicole-weinert-644885114/

Sumary

Today the Austrian AI Podcast is in Salzburg visiting the Salz21 an Innovation conference, that by the time this episode is out, has taken place from the 15 to the 16 of march 2023 in the city of Salzburg. This year there is a special interest in Artificial Intelligence, dedicating the "Stage of AI" to different presentations and discussion with a fast selection of speakers and topics.

I had the pleasure to meet several of the speakers and to record in total 3 interviews with the followings guests

Wolfgang Trutschnig from the University of Salzburg on their AI bachelor and Master Program, as well on IDALab, an applied AI research lab that collaborates closely with companies in the region to enable funding of basic research projects and support local companies to stay ahead of their competitors.
Gabrielle Bolek Fügl on a Salz21 panel discussion on Trustworthy AI and a workshop program that she has developed with Carina Zehetmaier that helps AI companies to develop products and services that are aligned with regulations and legal requirements.
And last but not least, Mario Tuta from Stoic Analytics on what holds AI adoption back for small and medium companies in austria, and what is needed to overcome those hurdles.

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

References

https://www.linkedin.com/in/mariotuta/ - Mario Tuta

KMU Digitalisierungsstudie 2019 - Digitale Transformation von KMUs in Österreich by Arthur D. Little

https://stoic-analytics.com/

TOC

00:00:00 Introduction

00:01:29 Beginning of the Interview

00:04:26 Career as a Data Science & Data Engineer Team Lead at Swarovski

00:12:17 First use cases and challenges on an internal cloud based data and compute platform

00:17:07 Salz21 Panel Discussion about challenges to adapt AI in SME's in Austria

00:24:05 How SME's identify AI services and solutions for their use cases

00:29:45 What is needed within a company to ease AI adoption

00:31:57 Buy or build?

00:38:23 How to grow AI adoptions within a company

00:42:46 The impact of Generative AI on SME's

00:47:08 How much will generative AI replace jobs or boost the productivity of people?

Sumary

Today the Austrian AI Podcast is in Salzburg visiting the Salz21 an Innovation conference, that by the time this episode is out, has taken place from the 15 to the 16 of march 2023 in the city of Salzburg. This year there is a special interest in Artificial Intelligence, dedicating the "Stage of AI" to different presentations and discussion with a fast selection of speakers and topics.

I had the pleasure to meet several of the speakers and to record in total 3 interviews with the followings guests

Wolfgang Trutschnig from the University of Salzburg on their AI bachelor and Master Program, as well on IDALab, an applied AI research lab that collaborates closely with companies in the region to enable funding of basic research projects and support local companies to stay ahead of their competitors.
Gabrielle Bolek Fügl on a Salz21 panel discussion on Trustworthy AI and a workshop program that she has developed with Carina Zehetmaier that helps AI companies to develop products and services that are aligned with regulations and legal requirements.
And last but not least, Mario Tuta from Stoic Analytics on what holds AI adoption back for small and medium companies in austria, and what is needed to overcome those hurdles.

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

References

https://www.linkedin.com/in/gabriele-bolek-f%C3%BCgl/ - Gabriele Bolek-Fügl

https://www.womeninai.at/ - Women in AI Austria

TOC

00:00:00 Introduction

00:01:29 Beginning of the Interview

00:03:44 Salz21 Pan el Discussion about Trustworthy AI Systems

00:07:42 How to integrate an AI System into a company

00:10:27 Workshop on building compliant AI Systems

00:14:44 Legal requirements for AI compliance

00:19:35 What companies benefit from an AI compliance workshop

00:24:55 Future Trustworthy AI Guidelines and Certifications

00:30:18 How to check and monitor AI Systems within a company

00:32:23 Biggest challenges when aligning AI systems with compliance requirements

00:43:10 Impact of AI on society and the labour market

Summary

Today the Austrian AI Podcast is in Salzburg visiting the Salz21 an Innovation conference, that by the time this episode is out, has taken place from the 15 to the 16 of march 2023 in the city of Salzburg. This year there is a special interest in Artificial Intelligence, dedicating the "Stage of AI" to different presentations and discussion with a fast selection of speakers and topics.

I had the pleasure to meet several of the speakers and to record in total 3 interviews with the followings guests

Wolfgang Trutschnig from the University of Salzburg on their AI bachelor and Master Program, as well on IDALab, an applied AI research lab that collaborates closely with companies in the region to enable funding of basic research projects and support local companies to stay ahead of their competitors.
Gabrielle Bolek on a Salz21 panel discussion on Trustworthy AI and a workshop program that she has developed with Carina Zehetmaier that helps AI companies to develop products and services that are aligned with regulations and legal requirements.
And last but not least, Mario Tuta from Stoic Analytics on what holds AI adoption back for small and medium companies in austria, and what is needed to overcome those hurdles.

Sponsors

Quantics: Supply Chain Planning for the new normal - the never normal - https://quantics.io/

References

Wolfgang Trutschnig - https://www.linkedin.com/in/wolfgang-trutschnig-a0527281/

IDALab - https://www.plus.ac.at/aihi/der-fachbereich/ida-lab/projekte/

Dr. Ulrike Ruprecht (Lab-Manager), email: ulrike.ruprecht@plus.ac.at

TOC

00:00:00 Introduction

00:01:28 Beginning of the Interview

00:04:03 AI Stage Panel Discussion about Dave an AI Service Agent

00:11:43 Uni Salzburg Master in Data Science and Bachelor in AI

00:18:00 IDALab

00:20:13 Projects and Collaborations at IDALab

00:24:13 The role of Basic AI Research in the near future

Summary

Today on the show I have the pleasure to talk to Elma Dervic. PhD at the Medical University of Vienna where she is applying network science to study patience medical records to identify how correlating deceases evolve over time and individual factors and events contribute to send patience on very different decease progressions. So doctors get a better understanding what treatments to apply in younger patients to reduce the probability of follow up complications when those patients get older; several decades into the future.

TOC

00:00:00 Start

00:01:40 Introduction

00:04:35 PhD in network science

00:08:13 The Medical Dataset used to study decease trajectories

00:18:18 Building network representations of medical data

00:32:27 Unsupervised clustering to identify communities of deceases

00:39:03 Studying and drawing conclusion from identified communities

00:47:23 Next steps and future work

References

https://www.linkedin.com/in/elmahot/ - Elma Hot Dervic

https://www.csh.ac.at/researcher/elma-dervic/ - Complexity Science Hub

https://www.linkedin.com/company/montenegrin-ai-association/ - MAIA

Summary

I must admit that I was thinking about starting my own startup for a long time, but never had the guts to really do so. One excuse that I always told myself was, that I am all alone, and I dont have a team with which to start that adventure.

Well if you are like me, than this episode is for you!

Today on the show I am talking to Adrian Spartarus. Founder of two AI Startups, and he did so as a single solo founder.

On this episode we will discuss some of the challenges that one faces as a solo founder and about building a AI product today.

We discuss the role of skill development, outsourcing, the importance of having the right training data and much more.

We end the episode by talking about the possibilities and dangers of the most recent AI advancements on society as a hole, but on modern tech Entrepreneurship in particular.

I hope you enjoy this episode and it will encourage you to take that step towards realizing your own dreams despite the difficulties ahead.

References

Adrian Spartaru - https://www.linkedin.com/in/spataru/

Cleanvoice - https://cleanvoice.ai/

Ramen Club - https://www.ramenclub.so/

Indi Hackers - https://www.indiehackers.com/

Indi world wide - https://indieworldwide.com/

Company of One: Why Staying Small Is the Next Big Thing for Business by Paul Jarvis

The Mom Test: How to talk to customers & learn if your business is a good idea when everyone is lying to you by Rob Fitzpatrick

Zero to Sold: How to Start, Run, and Sell a Bootstrapped Business by Arvin Cult

Summary

Today on the show I am talking to Kathrin Kefer PhD student at the JKU and her research at Fronius where she is optimizing single household power grids with their own photovoltaik installations to reduce costs and their dependency on the energy suppliers.

Kathrin will explain to us some of the challenges when controlling how power is used, stored and distributed in a system, as well how she has developed a control system that is applying symbolic regression in a genetic algorithm paradigm to find a balance between optimal control and realtime requirements.

References

Kathrin Kefer - https://www.linkedin.com/in/kathrin-kefer-362443132/

Symbolic regression: https://en.wikipedia.org/wiki/Symbolic_regression

Genetic Programming: https://en.wikipedia.org/wiki/Genetic_programming

Summary

Building AI solutions is hard. Using the AI hammer to be solving the actual source of a problem even harder.

Today on the show I am talking to Paul Puntschart on how to make sure to ask the right questions and try to use AI to answer the best questions that will solve the challenges that you are facing.

Paul will describe his approach of mixed intelligence. AI prototyping and playscience in order to through an playful iterative development process learn more about the problem you try to solve, learn how to ask the right questions and inspire creative solutions that hit the nail on the head.

References

Paul Puntschart : https://www.linkedin.com/in/paul-puntschart-279506a2/

NetHack Challenge: https://www.linkedin.com/pulse/nethack-challenge-i-participated-paul-puntschart/

Mixed Intelligence and Rapid AI Prototyping: https://www.linkedin.com/pulse/mixed-intelligence-rapid-ai-prototyping-paul-puntschart%3FtrackingId=8KRSQ3BYQP%252BSYzOrih0yiQ%253D%253D/?trackingId=8KRSQ3BYQP%2BSYzOrih0yiQ%3D%3D

Play: How it Shapes the Brain, Opens the Imagination, and Invigorates the Soul by Stuart Brown

Summary

Today on the show my guest is Mykola Belich found of Belichberg a software development company developing multiple products and services in the data and ML space. On the show Mykola will talk about their work in the area of video surveillance for photovoltaic plants in isolated rural areas.=

On the show he will share some of the biggest challenges that they had to overcome to apply computer vision on low power edge devices in deployment scenarios where maintenance is only possible under high costs or not at all.

References

Mykola & Belichberg - https://www.linkedin.com/feed/update/urn:li:activity:6997863170191073280/

Intels OpenVino - https://www.intel.com/content/www/us/en/developer/tools/openvino-toolkit/overview.html

Intro

Self driving cars have been a highly debated topic for years now, but today on the show we are not going to focus on the state of those autonomous vehicles, driver assisting systems or what so ever, but instead we are going to focus on the testing aspect of autonomous operating agents.

Today on the show I have the pleasure to talk to Florian Thaler, researcher at the Virtual Vehicle Research GmbH in Graz, focusing on testing reinforcement learning agents for safety critical systems.

Florian will tell us about his experience building a testing infrastructure for a vehicle development platform that applies the training paradigm of reinforcement learning to achieve its goals. He will share with us some of the important concepts and practises of testing RL and Machine Learning systems in general.

References

Florian Thaler: https://www.linkedin.com/in/florian-thaler-78744b214/
Virtual Vehicle Research Center: https://www.v2c2.at/
Spider Platform: https://www.youtube.com/watch?v=N7JgCbBiPtk
Spider Platform https://www.youtube.com/watch?v=5LDrxTxPDB8
https://arxiv.org/pdf/1906.10742.pdf : Testing Machine learning systems

Summary

Today on the Austrian AI Podcast, we join the hype and talk about big language models and program synthesis.

I am sure you have heard about OpenAI's code pilot, or AWS Codewhisper, or even about Google's internal projects to use big language models to help developers write better code, faster.

Program synthesis models are all over the news and today I have the pleasure to talk to Klaudius Kalcher cofounder of Magic.dev an AI startup that builds Software, that builds Software.

In the episode we talk about how to use big language models for program synthesis. About what is out there already, open challenges and what is the state of the art. Klaudius shares with us how he thinks about the program synthesis and how at Magic.dev they are building a software development support system in the form of an AI companion.

References

https://www.linkedin.com/in/klaudiuskalcher/

https://magic.dev/

https://arxiv.org/pdf/2207.11280.pdf - PanGu-Coder

https://github.com/features/copilot

https://aws.amazon.com/blogs/aws/now-in-preview-amazon-codewhisperer-ml-powered-coding-companion/

https://ai.googleblog.com/2022/07/ml-enhanced-code-completion-improves.html

Summary

Today on the show we will be talking about agent based modelling that is used to simulate macro economical systems, like the British labour market. For this I am talking to Maria del Rio Chanona currently doing her PhD at the Complexity Science Hub Vienna.

Maria will not only tell us about ways to use agent based modelling to simulate the effect of policy changes or external events like the pandamic on complex systems like the labour market, but during our deep dive, she will talk about her most recent work that focuses on using NLP to tab new sources of fine grain data on an individual level, like social media that can be used feed the next generation of agent based models.

References

https://www.csh.ac.at/researcher/maria-del_rio-chanona/

https://mariadelriochanona.info/

Forecasting the propagation of pandemic shocks with a dynamic input-output model https://www.sciencedirect.com/science/article/pii/S0165188922002317

Occupational mobility and automation: a data-driven network model https://royalsocietypublishing.org/doi/full/10.1098/rsif.2020.0898

Mental health concerns prelude the Great Resignation: Evidence from Social Media https://arxiv.org/abs/2208.07926

Episode

Most AI practitioner's, including myself, think of long running algorithms as something that is caused by big data or poor implementation and that can be solved best by more compute, but today on the show we will be discussing hard problems and their runtime complexity with Kirill Simonov from the algorithm and complexity group at the technical university Vienna.

Kirill is talking about this research in algorithm complexity and gives us a taste of how to solve hard problems with for example, approximation algorithms, that exchanging the accuracy or correctness of results for lower runtimes, or parameterized complexity algorithms that reduce runtime by limiting the solution space.

References

Kirill Simonov: https://www.ac.tuwien.ac.at/people/ksimonov/

Thesis: https://bora.uib.no/bora-xmlui/bitstream/handle/11250/2735169/archive.pdf?sequence=1&isAllowed=y

Lecture on Fixed-Parameter Algorithms: https://www.youtube.com/watch?v=4q-jmGrmxKs

Intro I am sure most of you are listening or looking at some weather forecast during the day and more often than we like to see, we read news about climate change causing new temperature records or glaciers melting at accelerating rates. Today we are not going to talk about climate change or weather forecast directly, but its underlying principle, Hydrology (which is study of water movement and distribution in a physical system). We will talk about strategies to build Hydrological models and more concretely we are looking at the intersection of Machine Learning and Hydrology. For this I am talking to Moritz Feigl. Co-founder and Chief Data Scientist at Baseflow.ai

During his PhD, Moritz investigated how Hydrology can benefit from Machine Learning, and in the interview we are going to contrast and compare two main approaches in Hydrological modeling. On one side we look at process-based models that are build on a systematic understanding of the physical world and principles and on the other side, at data-driven models; like modern Deep Learning systems that are learning a input-output relationship based on observations alone.

Moritz explains different ways how to combine those traditionally opposing approaches to get the best of both worlds, increasing the accuracy of predictions and enhancing our understanding of the underlying physical systems.

References

https://www.linkedin.com/in/moritz-feigl/

https://www.linkedin.com/company/baseflow-ai-solutions/

https://baseflow.ai/

https://abstracts.boku.ac.at/search_abstract.php?paID=3&paSID=19947&paSF=&paLIST=0&language_id=DE

Summary

Have you every had troubles paying your bills and got some nasty calls or letters about it? Depth collection is surely one area where I would not have thought to find AI, but today on the show, I am talking to Stephan Stricker Founder and CEO of Pair Finance and Maxime Kaniewicz, Data Science Team lead.

On how they combine the insights and methods from targeted marketing with reinforcement learning, to nudge customers towards paying their bills.

I think this episode is of great value to anyone who is thinking of building reinforcement learning systems for real business cases.

We speak about many of the main challenges in reinforcement learning, like how to collect intermediate rewards that match the business objects without running into the alignment problem. Or how to evaluate and compare different agents and policies without loosing revenue and cause damage to the business. We discuss the necessity of historical training data and the continuous flow of new training data in order to improve and optimize the system. We hear about ways to overcome the cold start problem by helping the agent to expand into new environments by providing new actions in combination with new priors and experiences.

I hope you will like this episode, and I can ensure you that there is a lot to learn.

References

https://www.pairfinance.com/

https://www.linkedin.com/in/stephanstricker/- Stephan Stricker - Founder and CEO of Pair Finance

https://www.linkedin.com/in/maxime-kaniewicz/- Maxime Kaniewicz - Data Science Team Lead

When we talk about data on this podcast, its mostly about training data and its properties that are relevant for the training machine learning models. But today we look at the bigger picture and the use of data in future data economies. How should the future use of data on a bigger scale look like? How can we make sure to build trustworthy and ethical data economy that follow our European Values?

Today on the show I am talking to Nina Popanton from the Data Intelligence Initiative (DIO) about its role as an enable of data collaborations. We talk about the challenges and the opportunities they see for companies, academia and the public sector when sharing data for specific use cases.

We discuss the motivation for stockholders to come together and share their data, and under which circumstances they are willing to do so.

In addition we are taking a step back and have a look the greater picture and the socioeconomic responsibility of a data sharing economy, driven by the "European Strategy for Data" and European projects like Gaia-X.

I know this episode diverges from our usual focus on AI and its methods, but I hope it will be an inspiration to you. Let's get started …

References

Nina Popanton - https://www.linkedin.com/in/nina-popanton-4b1541179/

DIO - https://www.dataintelligence.at/

Gaia-X - https://www.gaia-x.eu/

GreenData Hub - https://www.greendatahub.at/

Today on the show I am talking to Adrian Schiegl, the head of data science at XUND; an Austrian AI Startup that develops systems to predict medical diagnoses based on patients self reported symptoms.

During the interview Adrian is going to share some of the findings, challenges and solutions XUND has experienced and developed since its inception in 2018. For example, Xund's decision to move away from developing an mobile phone based self diagnosis system towards an Medical API that enables other vendors to integrate their automatic diagnoses system into their own products. In addition Adrian is telling us about recent research projects and future goals of the company, to move into hospitals and clinics in order to support the digitalization of a patients medical journey and ensure the most effective treatment possible.

References

Adrian Schiegl : https://www.linkedin.com/in/adrian-schiegl/

Xund: https://xund.ai/

Bayesian Neural Networks : https://proceedings.neurips.cc/paper/2020/hash/322f62469c5e3c7dc3e58f5a4d1ea399-Abstract.html

This is the second part of my interview with Hamid Eghbal-zadeh, post-doc at the Johannes Kepler University at the Institute of Machine Learning.

In the interview, we are talking about his research on a series of different aspects of representation learning with deep neural networks in order to make them more robust and improve their out-of-distribution behavior.

In this second part, we are talking about disentangled representations and the benefit they bring to agents trained in contextualized reinforcement tasks, in order to operate in unseen contexts and environments.

References:

Personal Homepage: https://eghbalz.github.io/
Hamid on LinkedIn: https://www.linkedin.com/in/hamid-eghbal-zadeh-8642b3a8/
H. Eghbal-zadeh, Representation Learning and Inference from Signals and Sequences, PhD Thesis, 2019.
H. Eghbal-zadeh, F. Henkel, G. Widmer, Context-Adaptive Reinforcement Learning using Unsupervised Learning of Context Variables, In Proceedings of Machine Learning Research, NeurIPS 2020 Workshop on Pre-registration in Machine Learning, PMLR 148:236-254, 2021.

This is the first part of my interview with Hamid Eghbal-zadeh, post-doc at the Johannes Kepler University at the Institute of Machine Learning.

In the interview, we are talking about his research on a series of different aspects of representation learning with deep neural networks in order to make them more robust and improve their out-of-distribution behavior.

In this first part, we are talking about the origin of representation learning and data augmentation. Hamid explains his research on the effects of representation learning on model training and highlights some of the important caveats that data augmentation can have on the robustness of your models.

References:

Personal Homepage: https://eghbalz.github.io/
Hamid on LinkedIn: https://www.linkedin.com/in/hamid-eghbal-zadeh-8642b3a8/
H. Eghbal-zadeh, Representation Learning and Inference from Signals and Sequences, PhD Thesis, 2019.
H. Eghbal-zadeh, F. Henkel, G. Widmer, Context-Adaptive Reinforcement Learning using Unsupervised Learning of Context Variables, In Proceedings of Machine Learning Research, NeurIPS 2020 Workshop on Pre-registration in Machine Learning, PMLR 148:236-254, 2021.

Show

Today on the show I have the pleasure to talk to Moritz Kampelmühler; Computer Vision and Machine Learning Research Engineer at Reactive Reality.

Reactive Reality is a spin-off from the Technical University of Graz that started out with developing a virtual dressing room that can be used by shoppers to create a virtual avatar of themselves and try out all kinds of different garments.

On the show, Moritz explains how they provide a rich shopper experience that scales making use of methods like photo-realistic rendering and state-of-the-art image processing in 2D and beyond.

In the second part of the Interview, we discuss the current research efforts at Reactive Reality that focus on 3D and how to create 3D reconstructions of people and garments from only a few or even a single picture.

References

Pictofit Shopping App: https://apps.apple.com/at/app/pictofit-shopping/id1527419372
Saito, S., Huang, Z., Natsume, R., Morishima, S., Kanazawa, A., & Li, H. (2019). Pifu: Pixel-aligned implicit function for high-resolution clothed human digitization. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 2304-2314).

Intro

Today on the show I have the pleasure to talk to Shiva Ghahremani, Co-founder and CEO of Asigmo Data Science.

Asigmo is a teaching company that has developed a data science Bootcamp to jump-start your career in Data Science.

In the interview, we talk about the structure of the Bootcamp and the content of the syllabus. We further discuss the role of boot camps in the wide area of online education and how at Asigmo they strive to find the perfect balance between the time students invest in the course and the skills they learn in return.

I think this episode is most interesting for people that are curious to find effective ways to pivot their career towards Data Science and alike. People with a background in software development that want to specialize or people in academia that are searching for a smooth transition into industry.

More than anything else, I hope you find this episode interesting and entertaining.

References

https://www.linkedin.com/in/shiva-ghahremani/

https://www.asigmo.com/

Summary

Yasin is a first year PhD Student working on improving information retrieval systems as part the European DOSSIER project. Where he is investigating new ways to improve the relevance of search results presented by interactive learning systems.

In particular his work focuses on ways to model, leverage and measure the user's Knowledge Gap; which is the difference between a user's prior understanding about a topic, before learning and their understanding after they have spend time using a learning systems to investigate a domain.

With his work, Yasin aspires to build the next generation of learning systems that reduce the time and afford users need to acquire knowledge. Building an adaptive system that is aware of a users understanding of a topic and can therefore adapt its response to present a user with the best next stepping stone on their learning journey.

As part of the interview Yasin is describing his main research topics and the progress so far.

In addition we are talking about different general aspects of Information retrieval systems, like user context and personalization.

References

Yasin Ghafourian - https://www.linkedin.com/in/yasinghafourian/
TU Wien, Ecommerce Group (https://informatics.tuwien.ac.at/orgs/e194-04)
Research Studio Austria (https://www.researchstudio.at/)
Dossier Project: https://dossier-project.eu/
Boughareb, D., & Farah, N. (2014, November). Context in information retrieval. In 2014 International Conference on Control, Decision and Information Technologies (CoDIT) (pp. 589-594). IEEE.

Intro

Carina Zehetmaier, CEO and Co-Founder of Taxtastic an Austrian AI Startup that develops small business and consumer solutions for income tax declaration.

We cover multiple topics on today's show. Starting out with talking about the technical challenges, Carina and her Team are facing when digitizing receipts. Not only how they make use of methods from image analysis in order to extract the relevant information from receipts, but as well how they classify receipts based on methods from NLP in order to deceit if a receipt is tax relevant or not.

What I think makes this episode special is our discussion about the responsibilities companies have when designing end-user products that make use of privacy-relevant customer data. Those companies should form the start designing their products and services with privacy in mind, and how they have to think about the user data they collect.

In addition, we talk about the difficulty to develop AI-powered end-user products that customers can trust, while still being aware of their limitations and imperfections.

We end the interview with Carina describing her work at Women in AI Austria to increase diversity in the field of AI and tech in general.

References

Carina Zehetmaier : https://www.linkedin.com/in/carina-zehetmaier/

Taxtastic : https://taxtastic.at/

Woman in AI : https://www.womeninai.at/

Intro

Michael is the co-founder and director of SO Digital Recruitment focusing on data positions in the DACH region, with a special focus on Austria.

During the Interview, Michael is looking back at 2021 from a recruiter's perspective on what skills and profiles have been in demand, and what is in decline. We are talking about the long-term effect of COVID on companies hiring strategies and how they need to adapt in order to continue to attract talent.

We end the interview by discussing the position Vienna has in the European AI Startup scene and how I can become more prominent.

References

Michael Outar - https://www.linkedin.com/in/michaeloutar/

SO Digital Recruitment - https://www.linkedin.com/company/so-digital-recruitment/

Intro

Resul Akay is the Chief Data Scientist at Quantics.io, where he is developing a platform to enable predictive modeling for supply chain optimization. Enabling customers from different industries and different steps of their supply chain, to accurately predict demand and optimize their supplies accordingly.

In the interview Resul explains to us some of the main challenges of supply chain optimization, and how at Quantics.io they address those challenges by making use of a self-aware computing paradigm to develop a fore casting engine that can analyze its prediction error and trace it to different actors and factors along the supply chain. Enabling the system to learn what to ignore and what to pay attention to over time to improve its forecasting accuracy.

References

Resul Akay - https://www.linkedin.com/in/resul-akay-5287061b3/

Quantics.io - https://quantics.io/

Self-aware Computing Systems, An Engineering Approach, Peter R. Lewis et.al - https://doi.org/10.1007/978-3-319-39675-0

Intro

Noah Weber is the Chief Technology offer at Celeris Therapeutics. Celeris is discovering candidates for targeted protein degradation in silico using Artificial Intelligence, to accelerate drug discovery and find cures for diseases like Alzheimer's.

In the first part of the show, he shares his experience on the importance of good MLOps practices and the importance of cloud solutions for the development of AI products at scale.

In the second part, we talk about the core technology that Celeris has developed to solve the problem of finding candidates for targeted protein degradation. Noah explains the problem they are solving from a computational perspective and we discuss how they use Bayesian optimization to evaluate and rank candidates in silico, reducing the costs and time for drug discovery.

References

Noah Weber - https://www.linkedin.com/in/noahweber123

Celeris Therapeutics - https://celeristx.com

AI-driven degrader design - https://www.youtube.com/watch?v=p2OjZ5afyLw

ACIT Podcast with Noah Weber - https://open.spotify.com/episode/61eXYwKpVuj1w8l0fiyEh5?si=n0vKpaXLTNKz0nwIpJaGMw

Taylor Peer is currently working as the Director of Data Science at Cortical.io, a Viennese AI company developing business solutions for information extraction and processing using Natural Language Understanding.

On the show we are talking about Semantic Folding, the core technology developed by Cortical.io that is enabling all their business products, and how the company evolved from a technology provider to a solution provider, and the challenges of integration data driven AI solutions at customers.

References

Taylor Peer - www.linkedin.com/in/taylorpeer/

Cotical.io - www.cortical.io/

White paper on Semantic Folding - www.cortical.io/static/downloads/semantic-folding-theory-white-paper.pdf

Intro

Jelena Milosevic is currently working as a Data Scientist focusing on the application of machine learning in embedded devices in industrial settings.

On the show we are going to talk about the challenges in developing ML models for mobile and Embedded devices like limited compute and power. In addition we are touching on several information security aspects for such ML use cases.

References

Jelena Milosevic - https://www.linkedin.com/in/milosevicjelena/

Paper: Time, accuracy and power consumption tradeoff in mobile malware detection systems - https://doi.org/10.1016/j.cose.2019.01.001

Intro

Marco Mondelli is a group leader at the IST Austrian, focusing on theoretical machine learning and in particular on properties and behaviour of gradient descent methods when used to train overparameterized deep neural networks.

In this interview Marco describes his reasons to start a theoretical machine learning research Group at the IST Austria and several aspects of the IST PhD program.

In the second part of the interview we discuss the research done in his groups and recent publications investigating the reasons behind the efficiency of gradient descent algorithms in optimising deep neural networks.

References

Marco Mondelli - http://marcomondelli.com/

Mondelli group at IST: https://ist.ac.at/en/research/mondelli-group/

Mean-field particle methods: https://en.wikipedia.org/wiki/Mean-field_particle_methods

Landscape connectivity and dropout stability of SGD solutions for overparameterized neural networks : https://research-explorer.app.ist.ac.at/record/9198

Interview Timings

03:30 Personal intro & career development

15:47 The Mondelli research group at the IST

32:00 Main research focus

39:00 Recent Publication on the connectivity of loss landscape

56:00 Future research interests

Julia is a researcher in the E-Commerce Group at the TU Vienna, focusing on the analysis of social networks.

In the first part of the interview, Julia will describe the motivations and possibilities of performing network analysis on social networks. She will give examples of such analysis from the literature, as well as from her own research.

In the second part of the interview, Julia will talk about the digital humanism initiative. A international collaboration of scholars, policy makers and industrial players with the goal to ensure that the digital transformation of our society is aligned with our human centered interests.

References

Julia Neidhardt : https://informatics.tuwien.ac.at/people/julia-neidhardt

Digital Humanism Initiative : https://dighum.ec.tuwien.ac.at/

Summary

Rania has a background in theoretical mathematics and has focused her work in recent years as a Data Scientists on natural language understanding and social media monitoring.

Today on the show she will share her experience of organising Data4Good, an NGO organisation that is bringing together DataScientists and other NGO's on a voluntary bases in order to support different social projects.

During the second part of the interview, Rania will describe the ongoing efforts by women in AI Austria, to discuss and improve future European AI regulations. She describes what options individual and small organisations have to change international regulations on topics of AI and digitalisation in order to build a better future.

References

Rania Wazir : https://www.linkedin.com/in/raniawazir/ , rania.wazir@vdsg.at
Data4good : https://vdsg.at/data4good/
Women in AI Discussion Group : https://www.meetup.com/es/Women-in-AI-Austria/
Access now (defend digital rights online) : https://www.accessnow.org/
European digital rights : https://edri.org/
Algorithm Watchers : https://algorithmwatch.org/de/
European data protection supervisor : https://edps.europa.eu/_en
Regulatory frame for AI by the European parliament : https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
Coded Bias (Movie): https://www.imdb.com/title/tt11394170/

Summary

Christoph is the Chief Operation Officer and one of the co-founders of Image Biopsy Labs. IB Labs, is a Viennese startup that is developing a modular AI platform to accelerate the work of radiologists ensuring high precision, transparency and control.

On the interview, Christoph is sharing his experience on the most important aspects of creating an AI startup for safety critical applications. What stages such a startup has to pass in order to be successful, and some of the biggest challenges that it needs to overcome.

References

Christoph Götz : https://www.linkedin.com/in/chgtz/

Private Homepage http://www.christophgoetz.com/

Image biopsy lab : https://imagebiopsy.com/

Best new radiology vendor (bottom of page) : https://www.auntminnieeurope.com/index.aspx?sec=sup&sub=cto&pag=dis&ItemID=619843

Michael Freeman Award 2021 (third tab) : https://vec.efort.org/awards-winners-2021#t3

Product Demo : https://www.youtube.com/watch?v=MZ7MRLjB_QI&feature=emb_title

INiTS : http://www.inits.at/

Summary

In this episode, Linda Anderson a computational Linguistic by training and founder of the AI startup "Artificial Researcher" is talking about the history and challenges of domain specific text mining.

We are discussing why many of the typical DL driven approaches fail on domain specific text mining applications and how her company "Artificial Researcher" is coping with those challenges.

References

Linda Anderson : https://www.linkedin.com/in/linda-andersson-76483916/
Linda Anderson (TU Wien) http://ifs.tuwien.ac.at/~andersson/
Artificial Researcher : https://www.artificialresearcher.com/
Adversarial Examples in NLP : https://towardsdatascience.com/what-are-adversarial-examples-in-nlp-f928c574478e
D. Biber. Representativeness in corpus design. Literary and linguistic computing, 8(4):243–257, 1993
K. Erk. What do you know about an alligator when you know the company it keeps? Semantics and Pragmatics, 9(17):1–63, April 2016.
J.R. Firth. A synopsis of linguistic theory 1930-1955. Studies in linguistic analysis, pages 1–32, 1957
E. Keizer. The English Noun Phrase: The Nature of Linguistic Categorization. Studies in English Language. Cambridge University Press, 2007
G. Zipf. Human Behaviour and the Principle of Least-Effort. Addison Wesley, Cambridge, MA, 1949
C. K. Schultz, (editor) H. P. Luhn: pioneer of information science, selected works, New York, Spartan Books, 1968

In this episode, Frank is sharing his experience in teaching about AI over the years at different institutions to students from all backgrounds, ranging from businesses focused questions about digitalisation to CEO's, or guiding domain experts in their first experiences in areas of data science, to software developers about the implementation of ML algorithms and methods.

In the second part of the interview, we discuss the challenges and opportunities of the digital transformation for universities, and the role of AI for the future of education.

References:

Frank Benda : https://www.linkedin.com/in/frank-benda-phd-36a268147/

POBS: Private online business school - https://dba-fernstudium.eu/

MIT Technology Report: https://www.technologyreview.com/2019/08/02/131198/china-squirrel-has-started-a-grand-experiment-in-ai-education-it-could-reshape-how-the/

Squirrel AI : http://squirrelai.com/

Publications : https://link.springer.com/article/10.1007/s00291-019-00567-8

What kind of AI related services does azure cloud offer? What are some common reasons companies move into the cloud, and what are common first mistakes that companies encounter?

If those are questions that interest you, then the first part of the interview is for you!

In the second part of the interview, Sanja talks about the global initiative "Women in AI" that she as the Vienna Lead is supporting in order to help women in the field of AI to connect, support each other and increase the representation of female role models in tech.

Don't miss it!

References:

Sanja Jovanovic: https://www.linkedin.com/in/sanja-jovanovic/

Women in AI: https://www.womeninai.co/

What are the challenges hiring for an AI startups? How do you hire today, but be prepared for continously changing requirements of tomorrow? Or how do you convince the HR department of your talent and motivation?

If those questions have crossed your mint recently, than this episode with Tanja Zinkl, hiring for the successful Austrian AI company Kaleido, is for you!

References:

Tanja Zinkl : https://www.linkedin.com/in/tanja-zinkl-4aa647114/

Book : Data Teams - A unified Management Model for Successful Data-Focused-Teams by Jesse Anderson

Have you ever been excited to browse the other side of the world on google earth? Just to be disappointed by the poor quality of the 3d reconstructions, and the limited level of detail?

The Graz based AI company blackshark.ai has you covered!

In todays interview, Stefan Habenschuss, leading machine learning at blackshark.ai is explaining

their massive scalable aerial image segmentation system that has been used to create a digital

twin earth with more than 1.5 Billion 3d building reconstructions with an astonishing level of detail.

References:

Stefan Habenschuss : https://www.linkedin.com/in/stefan-habenschuss-708459102/

A twin earth for the Microsoft Flight simulator : https://www.youtube.com/watch?v=ng-mGNqLe6M

Massively Scalable Aerial Image Segmentation for Virtual Sensor Simulation Environments : https://lnkd.in/dafgdvm (Available after a free registration at NVIDIA's GTC event platform)

The field of AI and Machine learning is moving with an ever increasing pace, and for most small and medium companies that cannot afford their own research labs, keeping up new scientific papers and software frameworks is almost impossible.

Today on the show I interview Lukas Fischer who is the Research Manager of the Data Science branch of the Software Competence Center Hagenberg that is here to help companies achieve this difficult task. The center has the mission to enable the knowledge transfer from academia to private partners, so that they can stay ahead of the competition.

If you are interested to hear about the structure, challenges and career opportunities within an applied research institute at linking academia and private industry, then this episode is for you.

References:

Lukas Fischer: Lukas.Fischer@scch.at , https://www.linkedin.com/in/lukas-fischer-3015245b/

Software Competence Center Hagenberg : https://www.scch.at/en/news

SCCH COMET Modul S3AI: https://www.s3ai.at (Research in the area of Secure and Save Shared Artificial Intelligence Systems)

H2020 Project Teaming.ai: https://www.teamingai-project.eu/ (Human-AI-Teaming-Plattform for the development and improvement of AI systems in manufacturing)

H2020 Project Cogniplant: https://www.cogniplant-h2020.eu/ (Platform for process improvements in industrial settings)

H2020 Project SERUMS: https://www.serums-h2020.org/ (Privacy preserving ML in the medical domain)

FFG Project in cooperation with the St. Anna children cancer research VISIOMICS: www.visiomics.at

FFG COMET Funding: https://www.ffg.at/en/comet-competence-centers-excellent-technologies

SCCH SpinOff Sysparancy: https://www.sysparency.com/en/

On the show today, I have the pleasure to talk to Raphael Mitsch, currently working as a Senior Machine Learning Engineer at the Austrian AI consultancy enliteAI.

Are you curious to learn, what AI consultancy looks like in practice, and what skills are required to be successful as a consultant?

Or are you interested to hear about the challenges and opportunities of reinforcement learning in real world tasks?

If so, then this interview is for you!

Raphael Mitsch contact details

https://www.linkedin.com/in/raphaelmitsch/

r.mitsch@enlite.ai

github.com/rmitsch

Related links and references

https://github.com/enlite-ai/maze

https://www.enlite.ai/

Machine learning street talk #045 Microsoft's Platform for Reinforcement Learning (Bonsai): https://www.youtube.com/watch?v=zxs0wM-N1sQ

https://www.microsoft.com/en-us/ai/autonomous-systems-project-bonsai

https://gym.openai.com/

https://docs.ray.io/en/master/rllib.html

https://human-centered.ai/

On the show today, I have the pleasure to talk to Jillian Augustine.

Jillian is currently working as a Data Scientist in the Digital Excellence Team of the international paper and packaging manufacturer Mondi, but has received your Phd in Molecular Biology. On the show we are discussing Jillian experience working in academia and private industry. She is contrasting the mindset prominent on both sides, and what is needed to be successful, as well what industry can learn from academia, and vice versa.

In addition we briefly discuss different ways how domain experts and data scientists work together most efficiently, and how improvements of tooling and software might change this relationship in the future.

We end this episode, talking about diversity and inclusiveness. Jillian shares with us, what it means, and how it feels to be part of an underrepresented group. The importance of role models, and the intrinsic value of inclusiveness.

Jillian Augustine (LinkedIn) - https://www.linkedin.com/in/jillianaugustine/

Personal homepage - https://www.jill.codes/

Woman Techmakers Vienna : From Test Tube to Data Lakes by Jillian Augustine - https://www.youtube.com/watch?v=VFFU282Fkn8&t=39s

The MDN Web Docs Accessibility guide - This covers a range of best practices when working with HTML, JavaScript and CSS : https://developer.mozilla.org/en-US/docs/Web/Accessibility

Moving into Data Science from Non-CS Academia - Women in Data Science CEE 2021 : https://www.youtube.com/watch?v=HcvFKknw508

Summary

Today on the show, we have Frank Fichtenmueller.

Frank is currently working as a Senior Solution architect, at Crayon's AI Center of Excellence; here in Vienna, where he is developing distributed systems at scale that apply the power of Machine Learning to solve real world problems and enable new business cases.

Frank has been in AI, for more than a decade now, going through many ups and downs. He has build and sold his own startup, started, managed and grew AI teams from he ground up, as well as organised and led the the AI efforts and strategies of complete companies.

I believe this episode is of great value for anyone that either is in a position to build a new a AI team at his/her company or that want to understand how the interests of an AI engineering team can collide with business needs.

In addition Frank shares his experience on what skills and expertise new comers in the field are often missing and what he would recommend to focus on (Spoiler, its less fancy data science or AI but more solid software development and DevOp skills).

We end the episode discussion a possible future of Data Science and where the market is heading, away from research and development towards the integration and coordination of Machine Learning services.

I really hope you enjoy this episode and can benefit from the experience and wisdom Frank is so happily sharing with us.

References

Frank Fichtenmueller: https://www.linkedin.com/in/frankfichtenmueller/

Crayon: https://www.crayon.com/at/

Summary

On this episode of the Austrian AI Podcast, we have the pleasure to interview Ivona Vlčková, a young Data Scientist working at the Viennese AI startup Mostly.ai.

Ivona shares her experience working as a Data Scientists at a Startup with about 30 people and contrasts it with her experience working at an international company with thousands of employees.

We discuss her reasons to move to Vienna, and what Vienna has to offer and how it compares to other cities in attracting talent and companies in the AI sector.

Towards the end of the interview we discuss the topic of gender bias in Data Science and in Austrian companies in general.

It was a pleasure to host Ivona and a very interesting interview.

You can reach Ivona at ivona.vlckova@mostly.ai, or via LinkedIn at https://www.linkedin.com/in/ivona-vl%C4%8Dkov%C3%A1-7175b8ab/ .

References

Mostly.ai: https://mostly.ai/

Gender Bias in Computer Science Degree at the TU-Vienna: https://www.tuwien.at/tu-wien/ueber-die-tuw/zahlen-und-fakten#c9039

Book Recommendation: Jesse Anderson - Data Teams : https://www.datateams.io

On Differential Privacy: Differential Privacy:

A Primer for a Non-technical Audience, https://privacytools.seas.harvard.edu/files/privacytools/files/pedagogical-document-dp_new.pdf

Welcome to the Austrian AI Podcast!

I am your host, Manuel Pasieka and I am excited to share with you on this Podcast

my passion for Artificial Intelligence, Machine Learning and Data Science.

I will bring onto the show, experts from academia and privat industry, hobbyists

from none profit organisations, and decision makers in public institutions to talk

with them about the possibilities and potential of AI in Austria.

I hope you will join me for this journey, and you are as excited about it as I am.