Randomized multi-class classification under system constraints: a unified approach via post-processing

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chzhen, Evgenii, Hebiri, Mohamed, Taturyan, Gayane
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909965314162688
author Chzhen, Evgenii
Hebiri, Mohamed
Taturyan, Gayane
author_facet Chzhen, Evgenii
Hebiri, Mohamed
Taturyan, Gayane
contents We study the problem of multi-class classification under system-level constraints expressible as linear functionals over randomized classifiers. We propose a post-processing approach that adjusts a given base classifier to satisfy general constraints without retraining. Our method formulates the problem as a linearly constrained stochastic program over randomized classifiers, and leverages entropic regularization and dual optimization techniques to construct a feasible solution. We provide finite-sample guarantees for the risk and constraint satisfaction for the final output of our algorithm under minimal assumptions. The framework accommodates a broad class of constraints, including fairness, abstention, and churn requirements.
format Preprint
id arxiv_https___arxiv_org_abs_2512_14246
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Randomized multi-class classification under system constraints: a unified approach via post-processing
Chzhen, Evgenii
Hebiri, Mohamed
Taturyan, Gayane
Optimization and Control
Machine Learning
We study the problem of multi-class classification under system-level constraints expressible as linear functionals over randomized classifiers. We propose a post-processing approach that adjusts a given base classifier to satisfy general constraints without retraining. Our method formulates the problem as a linearly constrained stochastic program over randomized classifiers, and leverages entropic regularization and dual optimization techniques to construct a feasible solution. We provide finite-sample guarantees for the risk and constraint satisfaction for the final output of our algorithm under minimal assumptions. The framework accommodates a broad class of constraints, including fairness, abstention, and churn requirements.
title Randomized multi-class classification under system constraints: a unified approach via post-processing
topic Optimization and Control
Machine Learning
url https://arxiv.org/abs/2512.14246