Mi Zhang, assistant professor of electrical and computer engineering, was invited by Google to discuss the future of federated learning during a special online event July 29-30. Google researchers and faculty from around the world came together virtually to collaborate and establish partnerships to advance research on federated learning.
Q: What is federated learning?
Federated learning is an emerging computing paradigm that enables multiple end devices, like mobile phones, to collaboratively learn a machine learning model while keeping all the training data on the device. Since each end device's raw data is stored locally and not exchanged or transferred, end device users can benefit from obtaining a well-trained machine learning model without sharing their privacy-sensitive data.
Q: Why are computing experts and researchers at Google interested in this new paradigm?
Today, gigantic amounts of data are generated by mobile phones on a daily basis. These data contain valuable information about users and their personal preferences. With federated learning, Google is able to extract such valuable information while preserving the privacy of user data to build better and personalized machine learning models to deliver individualized services to maximally enhance user experiences.
Q: How does federated learning tie into your current research?
Artificial intelligence is moving from the cloud to billions of end devices such as mobile phones and Internet of Things (IoT) that people cannot live without in their daily lives. My research aims to realize the vision of making AI ubiquitous. Federated learning perfectly aligns with such a vision.
Q: What’s next for your research?
Federated learning opens up a brand new research field in AI, and has a plethora of open problems and challenges to be solved. My students and I are developing techniques to improving efficiency and effectiveness of federated learning, enhancing robustness to adversarial attacks and failures, and exploring how to apply federated learning to emerging applications.