Theoretically Grounded Loss Functions and Algorithms for Score-Based Multi-Class Abstention

Anqi Mao, Mehryar Mohri, Yutao Zhong
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:4753-4761, 2024.

Abstract

Learning with abstention is a key scenario where the learner can abstain from making a prediction at some cost. In this paper, we analyze the score-based formulation of learning with abstention in the multi-class classification setting. We introduce new families of surrogate losses for the abstention loss function, which include the state-of-the-art surrogate losses in the single-stage setting and a novel family of loss functions in the two-stage setting. We prove strong non-asymptotic and hypothesis set-specific consistency guarantees for these surrogate losses, which upper-bound the estimation error of the abstention loss function in terms of the estimation error of the surrogate loss. Our bounds can help compare different score-based surrogates and guide the design of novel abstention algorithms by minimizing the proposed surrogate losses. We experimentally evaluate our new algorithms on CIFAR-10, CIFAR-100, and SVHN datasets and the practical significance of our new surrogate losses and two-stage abstention algorithms. Our results also show that the relative performance of the state-of-the-art score-based surrogate losses can vary across datasets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-mao24a, title = { Theoretically Grounded Loss Functions and Algorithms for Score-Based Multi-Class Abstention }, author = {Mao, Anqi and Mohri, Mehryar and Zhong, Yutao}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {4753--4761}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/mao24a/mao24a.pdf}, url = {https://proceedings.mlr.press/v238/mao24a.html}, abstract = { Learning with abstention is a key scenario where the learner can abstain from making a prediction at some cost. In this paper, we analyze the score-based formulation of learning with abstention in the multi-class classification setting. We introduce new families of surrogate losses for the abstention loss function, which include the state-of-the-art surrogate losses in the single-stage setting and a novel family of loss functions in the two-stage setting. We prove strong non-asymptotic and hypothesis set-specific consistency guarantees for these surrogate losses, which upper-bound the estimation error of the abstention loss function in terms of the estimation error of the surrogate loss. Our bounds can help compare different score-based surrogates and guide the design of novel abstention algorithms by minimizing the proposed surrogate losses. We experimentally evaluate our new algorithms on CIFAR-10, CIFAR-100, and SVHN datasets and the practical significance of our new surrogate losses and two-stage abstention algorithms. Our results also show that the relative performance of the state-of-the-art score-based surrogate losses can vary across datasets. } }
Endnote
%0 Conference Paper %T Theoretically Grounded Loss Functions and Algorithms for Score-Based Multi-Class Abstention %A Anqi Mao %A Mehryar Mohri %A Yutao Zhong %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-mao24a %I PMLR %P 4753--4761 %U https://proceedings.mlr.press/v238/mao24a.html %V 238 %X Learning with abstention is a key scenario where the learner can abstain from making a prediction at some cost. In this paper, we analyze the score-based formulation of learning with abstention in the multi-class classification setting. We introduce new families of surrogate losses for the abstention loss function, which include the state-of-the-art surrogate losses in the single-stage setting and a novel family of loss functions in the two-stage setting. We prove strong non-asymptotic and hypothesis set-specific consistency guarantees for these surrogate losses, which upper-bound the estimation error of the abstention loss function in terms of the estimation error of the surrogate loss. Our bounds can help compare different score-based surrogates and guide the design of novel abstention algorithms by minimizing the proposed surrogate losses. We experimentally evaluate our new algorithms on CIFAR-10, CIFAR-100, and SVHN datasets and the practical significance of our new surrogate losses and two-stage abstention algorithms. Our results also show that the relative performance of the state-of-the-art score-based surrogate losses can vary across datasets.
APA
Mao, A., Mohri, M. & Zhong, Y.. (2024). Theoretically Grounded Loss Functions and Algorithms for Score-Based Multi-Class Abstention . Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:4753-4761 Available from https://proceedings.mlr.press/v238/mao24a.html.

Related Material