Explore the challenges of robust machine learning in this 46-minute conference talk by Nikos Tsilivis from New York University. Delve into the performance discrepancies between neural networks in standard classification settings and worst-case scenarios. Examine the concept of implicit bias in optimization and its increased importance in robust settings. Investigate the reasons behind large generalization gaps and overfitting in robust Empirical Risk Minimization (ERM). Gain insights into the critical role of model specification, including optimization algorithms and architecture, in robust machine learning. Consider current and future challenges in the field as presented by Tsilivis at IPAM's Theory and Practice of Deep Learning Workshop.
The Price of Implicit Bias in Robust Machine Learning
Institute for Pure & Applied Mathematics (IPAM) via YouTube
Overview
Syllabus
Nikos Tsilivis - The Price of Implicit Bias in Robust ML - IPAM at UCLA
Taught by
Institute for Pure & Applied Mathematics (IPAM)
Related Courses
-
The Elusive Generalization and Easy Optimization in Machine Learning - Part 1
-
The Elusive Generalization and Easy Optimization - Part 2
-
Does Generalization Imply Accuracy on the Line? A New Look at Robust Generalization
-
Overfitting: Benign, Tempered and Harmful - Lecture on Machine Learning Regularization
-
Where Does Negative Transfer Come From? On the Implicit Bias of SGD in Multi-Task Learning
-
Generalization in Deep Learning Through the Lens of Implicit Rank Minimization
