跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.26075 · v1

Understanding MoE Decisions Beyond Gradients

首次提交 2026-11-19 21:16
版本更新 2026-11-19 21:16
雷达观测 2026-11-20 16:53
arXiv ↗PDF ↗

摘要

Understanding a language model's decision requires knowing how its internal computations affect its preference between answers. Gradient-based methods estimate this influence by predicting the effect of changing those computations. In mixture-of-experts (MoE) models, replacing one expert can change which experts later layers select. Those switches can alter the answer scores in ways that derivatives do not capture. We introduce a predictor that uses the model's own expert outputs and selection rules to estimate these extra effects, then adds them to a second-order prediction. Across questions from three benchmarks, our method reduces the error in predicting answer-score changes by 60\% on Qwen3.6 and 47\% on Gemma4, relative to a multistep baseline using gradients and second derivatives.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-19 21:16 · 最新观测版本

Understanding MoE Decisions Beyond Gradients

Understanding a language model's decision requires knowing how its internal computations affect its preference between answers. Gradient-based methods estimate this influence by predicting the effect of changing those computations. In mixture-of-experts (MoE) models, replacing one expert can change which experts later layers select. Those switches can alter the answer scores in ways that derivatives do not capture. We introduce a predictor that uses the model's own expert outputs and selection rules to estimate these extra effects, then adds them to a second-order prediction. Across questions from three benchmarks, our method reduces the error in predicting answer-score changes by 60\% on Qwen3.6 and 47\% on Gemma4, relative to a multistep baseline using gradients and second derivatives.

查看此版本解读 →查看此版本原文 ↗