arXiv cs.AI· Minghe Zhang, Xuetao Kang, Yifei Ming, Yanfei Chen, Han Yu, Yufan Zhuang, Chen-Yu Lee·· 10 小时前AI 评分40
GUARD:用遗憾引导稳定递归自我改进
GUARD: Stabilizing Recursive Self-Improvement with Regret Guidance
AI 导读
GUARD 是一个无需外部评分器的闭环框架,用于自主 AI 智能体的持续 harness 演化。它将任务分解为执行前自检,通过方向性遗憾引导更新,并用 Soft-HURT 门控确保收益大于回退。在记忆、技能演化、持续任务流和领域推理等基准上,GUARD 持续提升性能,避免灾难性遗忘和记忆污染,同时大幅降低部署 token 消耗。
正文 · 原文
Abstract:Autonomous AI agents pair a frozen base model with an editable harness of prompts, skills, tool rules, and memory. Recursive self-improvement (RSI) adapts this harness from experience without weight updates, yet self-editing without external graders often backfires: reflection and scalar scores conflate what an edit newly broke with what remains unsolved, causing local regressions, skill forgetting, and memory bloat. We introduce GUARD (Grader-free Unsupervised Environment Design [UED] with Adaptive Rule Distillation), a closed-loop framework for reference-free continual harness evolution. Our key insight is to decompose each task into pre-execution self-checks and guide harness updates via directional regret. Each round grades trajectories against these checks, pinning mastered rules while prioritizing newly broken behaviors, and selects among parallel candidate edits by critic-free group-relative fitness. A Soft-HURT gate then promotes an edit only when gains outweigh regressions, scoping specialized fixes to prevent over-triggering and guaranteeing non-regression on its gated buffer. At test time, one frozen harness runs in a single pass. Across benchmarks on memory, skill evolution, continual task streams, and domain reasoning, GUARD consistently improves performance without catastrophic forgetting or memory pollution while substantially cutting deployment token usage.
| Comments: | Equal contribution: Minghe Zhang and Xuetao Kang. 38 pages, 8 figures, 13 tables |
| Subjects: | Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2611.23476 [cs.AI] |
| (or arXiv:2611.23476v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.23476 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Minghe Zhang [view email]
[v1]
Tue, 17 Nov 2026 20:26:34 UTC (508 KB)
来源:arXiv cs.AI · arxiv.org