Abstract: In optimization, convergence rates and complexity are essential notions to characterize the performance of algorithms. In most cases, however, such results hold for an arbitrary function class with generic properties. On the other hand, each time we minimize a particular function which is not necessarily the worst-case function. Adaptive algorithms aim to harness the further structure of optimization problems to go beyond such worst-case analysis. In this talk, I will discuss some examples of such algorithms and show how the step sizes in these algorithms give rise to compelling practical performance for applications such as neural network training or min-max games.
Adaptive algorithms
23.05.2022 14:20 - 15:05
Organiser:
R. I. Boţ
Location: