Federated Learning: The Art of Creating Models without Exchanging Any Training Data

 View Only
When:  Sep 15, 2022 from 02:00 PM to 03:00 PM (PT)
The popularity of machine learning models has dramatically increased in a large variety of applications that positively affect people’s daily lives, including product recommendations, healthcare predictions and critical applications. These models help augment systems’ abilities, leading to more productive environments. At the same time, building these models requires quality training data that is not always easy to acquire. Traditionally, all training data is transmitted to a central place to start the training. However, this practice is not always feasible. With new privacy legislation and regulation that inhibit free data sharing and transmission, producing models is becoming challenging. Federated learning has arisen as an alternative and privacy-friendly method, where multiple data owners collaborate to train a single model without exchanging any of their training data. This powerful new paradigm enables obtaining models that could not be created otherwise. For example, members of a consortium can now train models without fear of revealing their data to others.

In this talk, we will take a tour through the advantages, use cases and practical aspects of using federated learning in multiple scenarios. Then, I will explain how different use cases may require additional protections to achieve a desired privacy level. In particular, I will focus on how to decide what multi-party computation technique to select when applying federated learning in a diverse set of environments. Finally, I will highlight IBM Federated Learning, a framework that we have built for enterprise environments.

Speaker Bio

Nathalie Baracaldo leads the AI Security and Privacy Solutions team and is a Research Staff Member at IBM’s Almaden Research Center in San Jose, CA. Nathalie is passionate about delivering machine learning solutions that are highly accurate, withstand adversarial attacks and protect data privacy. Nathalie has led her team to the design of the IBM Federated Learning framework, which is now part of the Watson Machine Learning product. Nathalie is also the primary investigator for the DARPA program “Guaranteeing AI Robustness Against Deception” (GARD). In 2020, Nathalie received the IBM Master Inventor distinction for her contributions to IBM Intellectual Property and innovation. Nathalie also received the 2021 Corporate Technical Recognition, one of the highest recognitions provided to IBMers for breakthrough technical achievements that have led to notable market and industry success for IBM. This recognition was awarded for Nathalie's contribution to the Trusted AI Initiative. Nathalie co-edited the book “Federated Learning: A Comprehensive Overview of Methods and Applications”. Nathalie has received multiple best paper awards and published in top-tier conferences and journals. Nathalie received her Ph.D. degree from the University of Pittsburgh in 2016.