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 , and establish their realizable -consistency under mild conditions. For cost functions based on classification error, we further show that these losses admit -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 -consistency and Bayes-consistency of a specific surrogate loss. Furthermore, we identify choices of that lead to -consistent surrogate losses for any general cost function, thus achieving Bayes-consistency, realizable -consistency, and -consistency bounds simultaneously. We also investigate the relationship between -consistency bounds and realizable -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