252

Realizable HH-Consistent and Bayes-Consistent Loss Functions for Learning to Defer

Main:11 Pages
Bibliography:7 Pages
3 Tables
Appendix:16 Pages
Abstract

We present a comprehensive study of surrogate loss functions for learning to defer. We introduce a broad family of surrogate losses, parameterized by a non-increasing function Ψ\Psi, and establish their realizable HH-consistency under mild conditions. For cost functions based on classification error, we further show that these losses admit HH-consistency bounds when the hypothesis set is symmetric and complete, a property satisfied by common neural network and linear function hypothesis sets. Our results also resolve an open question raised in previous work (Mozannar et al., 2023) by proving the realizable HH-consistency and Bayes-consistency of a specific surrogate loss. Furthermore, we identify choices of Ψ\Psi that lead to HH-consistent surrogate losses for any general cost function, thus achieving Bayes-consistency, realizable HH-consistency, and HH-consistency bounds simultaneously. We also investigate the relationship between HH-consistency bounds and realizable HH-consistency in learning to defer, highlighting key differences from standard classification. Finally, we empirically evaluate our proposed surrogate losses and compare them with existing baselines.

View on arXiv
Comments on this paper