arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2510.10232cs.LGcs.AI
收藏

SGM:一种用于风险可控的递归自我修改的统计哥德尔机

SGM: A Statistical Godel Machine for Risk-Controlled Recursive Self-Modification

Fudan University(复旦大学);Tongji University(同济大学);University of California, San Diego(加州大学圣地亚哥分校)

Xuening Wu, Shenqin Yin, Yanlan Kang, Xinhang Zhang, Qianya Xu, Zeping Chen, Wenqiang Zhang

AI总结:

针对递归自我修改缺乏安全保障的问题,提出统计哥德尔机(SGM),用统计置信检验替代形式化证明,并引入确认触发谐波支出(CTHS)控制累积风险,在多个学习任务上验证了其有效性与鲁棒性。

摘要:

递归自我修改在自动机器学习、神经架构搜索和自适应优化中日益核心,然而现有框架均未确保此类修改的安全性。哥德尔机通过要求在重写代码前提供改进的形式化证明来提供原则性保障;然而,在随机、高维设置中此类证明难以获得。我们引入了统计哥德尔机(SGM),这是首个用于递归编辑的统计安全层。SGM 用统计置信检验(e值、霍夫丁界)取代基于证明的要求,仅在所选置信水平下认证优越性时才允许修改,同时分配全局错误预算以限制跨编辑的累积风险。我们还提出了确认触发的谐波支出(CTHS),该机制按确认事件而非轮次来索引支出,将错误预算集中于有前景的编辑,同时保持族系错误率。在监督学习、强化学习和黑盒优化上的实验验证了这一作用:SGM 在 CIFAR-100 上认证了真实收益,在 ImageNet-100 上拒绝了虚假改进,并在强化学习和优化任务上展示了鲁棒性。总体而言,这些结果将 SGM 定位为学习系统中持续、风险感知自我修改的基础设施。代码可在以下网址获取:此 https URL。

英文摘要:

Recursive self-modification is increasingly central in AutoML, neural architecture search, and adaptive optimization, yet no existing framework ensures that such changes are made safely. Godel machines offer a principled safeguard by requiring formal proofs of improvement before rewriting code; however, such proofs are unattainable in stochastic, high-dimensional settings. We introduce the Statistical Godel Machine (SGM), the first statistical safety layer for recursive edits. SGM replaces proof-based requirements with statistical confidence tests (e-values, Hoeffding bounds), admitting a modification only when superiority is certified at a chosen confidence level, while allocating a global error budget to bound cumulative risk across rounds.We also propose Confirm-Triggered Harmonic Spending (CTHS), which indexes spending by confirmation events rather than rounds, concentrating the error budget on promising edits while preserving familywise validity.Experiments across supervised learning, reinforcement learning, and black-box optimization validate this role: SGM certifies genuine gains on CIFAR-100, rejects spurious improvement on ImageNet-100, and demonstrates robustness on RL and optimization benchmarks.Together, these results position SGM as foundational infrastructure for continual, risk-aware self-modification in learning systems.Code is available at: https://github.com/gravitywavelet/sgm-anon.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑