A Theory of Multi-objective Machine Learning

Explore a comprehensive lecture on multi-objective machine learning presented by Nika Haghtalab from the University of California, Berkeley. Delve into the framework of multi-objective learning as a unifying paradigm for addressing robustness, collaboration, and fairness in machine learning. Discover how this approach aims to optimize complex and unstructured objectives using limited sampled data. Examine the relationship between multi-objective learning and classical and modern machine learning considerations, including generalization. Gain insights into technical tools with provable guarantees and review empirical evidence of their performance on important benchmarks. This talk, part of the Modern Paradigms in Generalization Boot Camp at the Simons Institute, offers a deep dive into the theory and practical applications of multi-objective machine learning.