English(EN) When a Flatness Proxy Is Not a Function: Robustness Certificates and Training Interventions

研究论文质疑用于机器学习模型鲁棒性的平坦度代理

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-02 04:00

一篇新发表在arXiv上的研究论文探讨了使用平坦度代理进行机器学习模型鲁棒性证书和训练干预的局限性。研究表明，一个常见的最后一层相对平坦度代理可以将损失增加低估210倍以上，并且softmax偏移会使代理变得无界。在CIFAR-10上的实验表明，这些干预措施可能导致泛化能力的实质性、可逆的抑制。 AI

影响强调了评估和改进模型鲁棒性常用方法的潜在缺陷，表明需要更可靠的技术。

排序理由研究论文发表在arXiv上，详细介绍了理论和实证发现。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.LG 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 13 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 研究论文发表在arXiv上，详细介绍了理论和实证发现。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.LG TIER_1 English(EN) · Vicente Opazo, Jose Calatayud-Mateu, Cristobal Rojas, Cristian Buc Calderon · 2026-10-02 04:00

当平坦度代理不是函数时：鲁棒性证书和训练干预

arXiv:2609.38540v1 Announce Type: new Abstract: A valid curvature upper bound need not justify either a robustness certificate or an intervention on an intrinsic predictor property. We demonstrate this distinction for a last-layer relative-flatness proxy used in both settings. Fi…

本文如何被排名

报道来源 [1]

当平坦度代理不是函数时：鲁棒性证书和训练干预

相关实体

相关话题