Skip to yearly menu bar Skip to main content


Poster

On Making Stochastic Classifiers Deterministic

Andrew Cotter · Maya Gupta · Harikrishna Narasimhan

East Exhibition Hall B + C #231

Keywords: [ Learning Theory ] [ Algorithms -> Classification; Theory ] [ Applications ] [ Fairness, Accountability, and Transparency ]


Abstract:

Stochastic classifiers arise in a number of machine learning problems, and have become especially prominent of late, as they often result from constrained optimization problems, e.g. for fairness, churn, or custom losses. Despite their utility, the inherent randomness of stochastic classifiers may cause them to be problematic to use in practice for a variety of practical reasons. In this paper, we attempt to answer the theoretical question of how well a stochastic classifier can be approximated by a deterministic one, and compare several different approaches, proving lower and upper bounds. We also experimentally investigate the pros and cons of these methods, not only in regard to how successfully each deterministic classifier approximates the original stochastic classifier, but also in terms of how well each addresses the other issues that can make stochastic classifiers undesirable.

Live content is unavailable. Log in and register to view live content