Konstantinos Daskalakis is a professor of Computer Science at ΜΙΤ. He is a graduate of the Department of Electrical and Computer Engineering at the National Technical University of Athens, did his PhD at the University of Berkeley, and worked as a postdoctoral researcher at Microsoft. His research focuses on theoretical computer science and its interface with Economics, Statistics and Artificial Intelligence. He has been awarded the Best Doctorate in Computer Science by the International Computer Science Organization (ACM), the Kalai Award from the International Game Theory Association, the Outstanding Publication Award from the International Association of Applied Mathematics SIAM, the Career Award from the National Science Foundation, the Sloan Foundation Computer Science Award, the Microsoft Research Fellowship and the top Rolf Nevanlinna Prize from the International Mathematical Society.
How does machine learning fail, and what to do about it?
As the applications of machine learning are exploding, it is also becoming increasingly clear that its use poses significant threats. Learning systems lack the type of robustness that one expects of systems that make critical decisions, failing to extrapolate well from their training to new environments, being extremely data and computation hungry, and amplifying biases in their training data. In this lecture I will look at some of the root causes making machine learning models fail, tinkering with the main assumptions of the machine learning pipeline, and revealing intimate connections to mathematics and the social sciences as potential avenues to overcome these issues.