Seminar Details

In this talk, we will explore the intersection between robust high-dimensional statistics and non-convex optimization. We will show that standard optimization methods such as gradient descent can efficiently solve various robust estimation tasks, and conversely, robust estimation algorithms can be used to develop robust algorithms for various tractable non-convex problems. Our results could lead to more practical and provably robust algorithms for many statistical and machine learning tasks, and shed light on the broader connections between robust estimation and non-convex optimization. This talk is based on joint work with Ilias Diakonikolas, Jelena Diakonikolas, Haichen Dong, Rong Ge, Shivam Gupta, Daniel Kane, Shuyao Li, Alessio Mazzetto, Mahdi Soltanolkotabi, and Stephen Wright.