Yang Liu
Assistant Professor, Computer Science and Engineering, University of California, Santa Cruz

Machine Learning (ML) is increasingly used in domains that have a profound effect on people's opportunities and well-being, including healthcare, law enforcement, and consumer finance. The effectiveness of ML depends on reliable datasets, training methods, and deployment, and additional challenges arise when each of the above components interacts with human agents. In this talk, I will detail some of my recent efforts on building a responsible and robust ML system with human in the loop. I will start with presenting a cohort of results on developing a new family of approaches to protect the training of ML models from potentially corrupted and biased training labels generated by human. Then I’d like to go beyond this static setting and discuss how to use the design and deployment of ML to offer human agents improvable actions to better develop their profiles and qualifications. My long-term goal is to build robust ML systems that promote healthy dynamics of the human-ML interaction.