Postdoctoral Researcher in Mathematical Machine Learning jobs

Adaptive Online Learning: Online learning deals with learning algorithms that process data sets sequentially, one data point at a time. Such algorithms play a key role in machine learning for very large data sets, for instance when training deep neural networks. The goal of this project is to develop new online learning algorithms and prove that they are optimal for multiple types of loss functions and under multiple different types of assumptions, without the need to manually tune any hyperparameters. Mathematical Formalization and Analysis of Explainability Methods: Recently there has been much interest in generating explanations that clarify and communicate how machine learning methods make their decisions. But none of the existing methods gives any formal mathematical guarantees on how well they can be expected to work and under which conditions their explanations can be trusted.