Abstract LLM alignment methods typically learn a single reward model (either implicitly or explicitly) from pairwise comparison data. This approach implicitly assumes homogeneous preferences across human labelers — an assumption that is violated in practice. As a result, the learned reward model is generally mis-specified: Prior work shows that it …
Classifiers are known to be vulnerable to adversarial examples, which are imperceptible modifications of true inputs that lead to misclassification. This raises many concerns, and recent research aims to better understand this phenomenon. We make progress on two fronts: 1) We take a holistic look at adversarial examples for non-parametric …
With a flurry of recent research motivated by applications to machine learning, convergence of gradient descent methods for smooth non-convex unconstrained optimization is well understood in the centralized setting. In this talk I will discuss our progress towards understanding how convergence of gradient descent methods (including SGD and acceleration) is …