Federated learning is relatively new, but it stands to have a huge impact on the machine learning landscape, especially as it applies to healthcare. On today’s episode of What That Means, Camille is joined by Olga Perepelkina, PhD, a Deep Learning Product Manager at Intel, to find out exactly how this field is projected to change the world.

They cover:

  • What federated learning is and why it matters

  • How it’s helping to protect sensitive data

  • How it differs from other machine learning approaches

  • The various ways federated learning can be used

  • How data is annotated in federated learning, and why it sometimes lacks quality control

  • Which components of the federated learning model are vulnerable to cyberattacks, as well as possible solutions to boost security

  • Whether or not you can have different kinds of input with the federated learning model

  • What the future holds for federated learning

...and more Great conversation, don’t miss it!

The views and opinions expressed are those of the guests and author and do not necessarily reflect the official policy or position of Intel Corporation.

Here are some key take-aways:

  • At less than five years old, federated learning is a completely new area of research in machine learning.

  • In federated learning, there is no need to collect data centrally; it can be trained locally, sending only model updates to an aggregation server.

  • It is crucial to protecting sensitive data because it keeps that data local on the devices where it was born.

  • Federated learning differs from other machine learning approaches in that models are trained in parallel, rather than sequentially.

  • In addition to medical imaging, federated learning can be used for things like computer vision applications, natural language processing (NLP), and deep learning.

  • One of the problems with federated learning is that the quality of data annotation cannot be directly observed; in the future, there will be some monitoring tools that will help to resolve this challenge.

  • Federated learning can help curb biased data sets because there is more diverse data involved.

  • In the future, as federated learning expands, communication efficiency will be key, as will the protection of models to prevent leakage of private information.

Some interesting quotes from today’s episode:

“I think we can dramatically improve AI in healthcare.”

“As researchers, we want to use these models to improve products, to build new technologies. But at the same time, we need to protect the private information of people, and federated learning can help to do that, to provide access to data, and to protect the privacy of people.”

“In federated learning, we keep data private on local devices where it was born. And we only send updates of the model to one server and aggregate this model, and then send this aggregated model back to local devices.”

“In federated learning, we still have some problems because people can't directly observe the quality of annotation. And this is one more major issue and challenge for federated learning.”

“In our future products, we will add some monitoring tools not to absorb your data, but to collect some statistics to help you in your research and your experiments.”

For more resources, check out these links:

OpenFL: Intel opensource federated learning library: https://github.com/intel/openfl

Federated Learning in Medicine: A Nature paper - https://www.nature.com/articles/s41598-020-69250-1

Intel Federated Learning Slack: https://join.slack.com/t/openfl/shared_invite/zt-ovzbohvn-T5fApk05~YS_iZhjJ5yaTw

See All 227 Episodes of "Cyber Security Inside"