Program error detection/correction: Turning PAC learning into Perfect learning
Computational learning theory is concerned with developing “scratch” a probably approximately correct (PAC) algorithm to solve a given computational problem. Program error detection /correction is concerned with transforming programs — such as these — which are correct on most instances into programs that are correct on all instances. The two approaches together enable one to generate a perfect program from scratch. The goal of this talk is to describe how this latter error detection/correction of algorithms works, and to encourage its integration into learning theory.