DeLTA seminar by Mladen Kolar
Speaker
Mladen Kolar, the University of Chicago Booth School of Business.
Title
An Adaptive Stochastic Sequential Quadratic Programming with Differentiable Exact Augmented Lagrangians
Abstract
In this talk, I will discuss our recent work on stochastic optimization withequality constraints. We consider solving nonlinear optimization problems withstochastic objective and deterministic equality constraints. We propose astochastic algorithm based on sequential quadratic programming (SQP) that uses adifferentiable exact augmented Lagrangian as the merit function. The design ofthe algorithm is motivated by an old SQP method (Lucidi, 1990) developed forsolving deterministic problems. I will first explain how to handle stochasticobjectives when the stepsizes are deterministic and prespecified. Next, I willexplain how to adaptively select the random stepsizes by adapting the stochasticline search procedure of Paquette and Scheinberg (2020) that was developed forunconstrained problems. We established the global ``almost sure" convergence forthe SQP method. If time permits, I will also discuss recent progress on solvingproblems with inequality constraints.
_____________________________
You can subscribe to the DeLTA Seminar mailing list by sending an empty email to delta-seminar-join@list.ku.dk.
Online calendar
DeLTA Lab page
DeLTA is a research group affiliated with the Department of Computer Science at the University of Copenhagen studying diverse aspects of Machine Learning Theory and its applications, including, but not limited to Reinforcement Learning, Online Learning and Bandits, PAC-Bayesian analysis