ABSTRACT
The availability of vast, easily accessible information has broadened our outlook on the world. Social networks helped us foster community interactions. Advances in machine learning have enabled tremendous progress in many complex tasks. At the same time, however, surveillance capitalism commodified personal information, and malevolent actors have tapped into electronic systems to erode citizens' privacy and attack people with whom they disagree. This has prompted a multitude of challenging research problems around understanding and countering privacy and security issues online.

In this talk, I will overview my research background and vision. I will focus on the work my team and I have done at the intersection of privacy and machine learning, presenting novel algorithms and systems to efficiently and securely train machine learning models and discussing their deployment in real-world applications. Then, I will focus on modeling and protecting privacy and safety in the context of deep neural networks and large language models vis-à-vis case studies like Federated Machine Learning and Chatbots. Finally, I will discuss a few open research problems in this space.