---
id: 20260710-T0-12
title: "MILES：模块化指令记忆让LLM在推理中持续自我改进"
title_en: "MILES: Modular Instruction Memory Helps LLMs Self-Improve During Reasoning"
url: https://ai.daily.yangsir.net/daily/20260710-T0-12
issue_date: 2026-07-10
publish_date: 2026-07-09T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2607.06974
---

# MILES：模块化指令记忆让LLM在推理中持续自我改进

arXiv 上新研究提出 MILES 框架，通过模块化指令记忆和可学习选择机制，让大语言模型在连续处理问题时积累并复用经验。与传统孤立处理每个问题的方式不同，MILES 允许模型自动选择有用的经验用于新问题推理，实现边推理边自我提升。实验显示该方法在多个推理基准上取得了显著效果。

## English Version

**MILES: Modular Instruction Memory Helps LLMs Self-Improve During Reasoning**

A new arXiv paper introduces MILES, a framework with modular instruction memory and learnable selection that enables LLMs to accumulate and reuse experience across sequential problems. Unlike isolated problem-solving, MILES allows models to autonomously select relevant past experiences for new tasks, achieving self-improvement during reasoning. Experiments show strong gains on multiple reasoning benchmarks.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2607.06974)

**详情页**：https://ai.daily.yangsir.net/daily/20260710-T0-12

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*