Robust Estimation via Multi-Output Optimization
Abstract
In this talk, we will discuss a general framework for addressing the problem of estimation in the presence of adversarial noise and outliers. Specifically, we consider the task of estimating a high-dimensional vector from a set of measurements that are subject to adversarial corruptions. We propose a robust estimator that is based on formulating and solving a multi-output optimization problem. The proposed estimator is efficient and provably recovers the true vector in the presence of a constant fraction of adversarial corruptions, under mild assumptions on the problem instance. We demonstrate the effectiveness of the proposed approach on various applications, including robust linear regression and matrix completion problems.
Bio
Amin Saberi is an Associate Professor of Management Science and Engineering at Stanford University. He received his Ph.D. in Computer Science from Stanford in 2005. His research interests lie in the areas of algorithms, optimization, and computational game theory. Saberi has made significant contributions to the field, and his work has been recognized with several awards, including the INFORMS Computing Society Prize for Operations Research and the Management Sciences, as well as the Erlang Prize from the INFORMS Applied Probability Society. He has also been a recipient of the National Science Foundation CAREER Award.
Additional Information
Amin Saberi is a co-founder of two startups: DFINITY, a blockchain platform focused on enabling decentralized applications, and Extend, a company that provides extended warranties for consumer electronics. His expertise in optimization and algorithm design has been influential in shaping the fields of computational economics, market design, and online advertising.
