Saved in:
Bibliographic Details
Main Authors: Mohri, Clara, Globerson, Amir, Kaplan, Haim, Koren, Tomer, Mansour, Yishay
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.28020
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916065968128000
author Mohri, Clara
Globerson, Amir
Kaplan, Haim
Koren, Tomer
Mansour, Yishay
author_facet Mohri, Clara
Globerson, Amir
Kaplan, Haim
Koren, Tomer
Mansour, Yishay
contents We consider the problem of Cost-Aware Learning, where sampling different components of a finite-sum objective incurs different costs. The objective is to reach a target error while minimizing the total cost. We propose Cost-Aware SGD, which uses a distribution based on gradient norms and costs to sample components. We provide a thorough analysis of this algorithm, including cost-improvement bounds over baselines, a characterization of distribution proxy sub-optimality, and a lower bound. We apply our theoretical insights to reinforcement learning with language models, where the computational cost of sequence-level policy gradients varies with length. We find that the advantage magnitude serves as a high-fidelity proxy for gradient norms, and use this to introduce Cost-Aware GRPO. Empirical results on 1.5B, 4B, and 8B LLMs demonstrate that this algorithm significantly reduces the tokens used in policy optimization while matching or exceeding baseline accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2604_28020
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Cost-Aware Learning
Mohri, Clara
Globerson, Amir
Kaplan, Haim
Koren, Tomer
Mansour, Yishay
Machine Learning
We consider the problem of Cost-Aware Learning, where sampling different components of a finite-sum objective incurs different costs. The objective is to reach a target error while minimizing the total cost. We propose Cost-Aware SGD, which uses a distribution based on gradient norms and costs to sample components. We provide a thorough analysis of this algorithm, including cost-improvement bounds over baselines, a characterization of distribution proxy sub-optimality, and a lower bound. We apply our theoretical insights to reinforcement learning with language models, where the computational cost of sequence-level policy gradients varies with length. We find that the advantage magnitude serves as a high-fidelity proxy for gradient norms, and use this to introduce Cost-Aware GRPO. Empirical results on 1.5B, 4B, and 8B LLMs demonstrate that this algorithm significantly reduces the tokens used in policy optimization while matching or exceeding baseline accuracy.
title Cost-Aware Learning
topic Machine Learning
url https://arxiv.org/abs/2604.28020