On the connection between compression learning and scenario based optimization

Abstract

We investigate the connections between compression learning and scenario based optimization. We first show how to strengthen, or relax the consistency assumption at the basis of compression learning and study the learning and generalization properties of the algorithm involved. We then consider different constrained optimization problems affected by uncertainty represented by means of scenarios. We show that the issue of providing guarantees on the probability of constraint violation reduces to a learning problem for an appropriately chosen algorithm that enjoys compression learning properties. The compression learning perspective provides a unifying framework for scenario based optimization and allows us to revisit the scenario approach and the probabilistically robust design, a recently developed technique based on a mixture of randomized and robust optimization, and to extend the guarantees on the probability of constraint violation to cascading optimization problems.