跳到正文
arXiv cs.AI· Behrooz Tahmasebi, Melanie Weber·· 10 小时前AI 评分34

Test-Time Symmetrization:面向结构下模型无关学习的缩放定律

Test-Time Symmetrization: Scaling Laws for Model-Agnostic Learning under Structure

AI 导读

论文提出 test-time symmetrization,通过对输入的重写(rewriting)做前向计算并按重写群 G 聚合答案,证明仅需 O(log |G|) 次随机重写即可恢复全部 |G| 次重写的信息,且该对数缩放为最优。

正文 · 原文

View PDF HTML (experimental)

Abstract:Test-time compute has emerged as a powerful way to improve the accuracy of language models at inference time, with scaling laws quantifying how accuracy improves with additional forward passes. Existing scaling laws, however, focus on repeated sampling from a fixed prompt, even though many tasks admit multiple equivalent representations of the same underlying problem. A graph does not depend on how its nodes are numbered, and a program does not depend on the names of its variables, yet language models can change their answers under such rewritings. In this paper, we study test-time symmetrization, which uses forward passes on rewritings of the input and aggregates the resulting answers over the group $G$ of rewritings. We prove that no amount of sampling and repeated decoding from a single representation can replace rewriting, while only $O(\log |G|)$ random rewritings suffice to recover the information provided by all $|G|$ rewritings. The same set of $O(\log |G|)$ rewritings provides simultaneous approximation across models, decoding rules, and sampled reasoning traces, and the logarithmic scaling is optimal in general. We validate the theory with experiments on Qwen2.5-7B, Phi-4-mini, Llama-3.1-8B, and Qwen2.5-14B, with chain-of-thought reasoning, on graph reasoning, logical deduction, and multiple-choice question answering. At the same number of forward passes, spending them on rewritings of the input is substantially more accurate than spending them on repeated sampling from the original prompt whenever the model's answer depends on how the input is written, and the number of rewritings needed grows only logarithmically with the number of possible rewritings, as the theory predicts. Our results provide a principled way to enforce symmetries at inference time when they cannot be built into the model, which we call model-agnostic symmetrization.
Comments: 38 pages, 13 figures
Subjects: Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.24032 [cs.AI]
  (or arXiv:2611.24032v1 [cs.AI] for this version)
  https://doi.org/10.48550/arXiv.2611.24032

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Behrooz Tahmasebi [view email]
[v1] Wed, 18 Nov 2026 08:20:34 UTC (258 KB)

来源:arXiv cs.AI · arxiv.org