[论文] Can Test-Time Memory Evolution Outperform Test-Time Training for Scien...

论文概要

研究领域: ML 作者: Yepeng Liu, Shengmai Chen, Xuandong Zhao, Chengzhi Liu, Dawn Song, Xin Eric Wang, Yuheng Bu 发布时间: 2026-10-27 arXiv: 2610.32046

中文摘要

最近的测试时训练（TTT）框架使开源大语言模型（LLM）在科学发现任务上超过了由前沿模型驱动的纯推理框架。然而，TTT在发现过程中引入了额外的在线权重更新计算成本，使得计算效率在扩展到更大模型时成为重要考虑因素。这促使我们探索一种高效的科学发现测试时自适应方法。我们引入测试时记忆演化（TTME），一种即插即用的测试时自适应方法，作为TTT的替代方案。TTME通过蒸馏经验持续更新外部记忆，以指导后续发现，而无需更新模型权重。为了进行干净比较，我们在相同的发现框架下评估TTME和TTT，仅替换测试时自适应方法。我们的评估表明，TTME在所有五项数学任务上都优于TTT，同时将计算成本在FLOPs上降低82.8-84.6%，在运行时间上降低47.7-58.4%。这些结果确立了TTME在所评估发现场景中作为TTT的有效且计算高效的替代方案。

原文摘要

Recent test-time training (TTT) frameworks have enabled open-weight large language models (LLMs) to outperform inference-only frameworks powered by frontier models on scientific discovery tasks. However, TTT introduces additional computational cost for online weight updates during the discovery process, making computational efficiency an important consideration when scaling to larger models. This motivates us to explore an efficient test-time adaptation method for scientific discovery. We introduce Test-Time Memory Evolution (TTME), a plug-and-play test-time adaptation method, as an alternative to TTT. TTME continually updates the external memory by distilling the experience to guide subsequent discovery without updating model weights. To have a clean comparison, we evaluate TTME and TTT u...

#论文 #arXiv #ML #小凯