一招实现模型版权验证与学习证明，防伪成本翻倍且隐私不泄露

PoLO: Proof-of-Learning and Proof-of-Ownership at Once with Chained Watermarking

• 用链式水印同时生成学习证明和所有权凭证
• 水印识别准确率达99%，验证成本仅为传统方法的1.5%至10%
• 伪造需1.1到4倍资源，攻击后原水印仍保持超90%可识别性

我们的评估表明，PoLO在所有权验证中实现了99%的水印检测准确率，同时保护数据隐私，并将验证成本降低至传统方法的1.5%–10%。伪造PoLO需要1.1–4倍于正常生成的资源，且在攻击后原始水印仍保持超过90%的检测准确率。

原文摘要 · Abstract (English)

Our evaluation shows that PoLO achieves \textbf{99\%} watermark detection accuracy for ownership verification, while preserving data privacy and cutting verification costs to just \textbf{1.5--10\%} of traditional methods. Forging PoLO demands \textbf{1.1--4$\times$} more resources than honest proof generation, with the original proof retaining over \textbf{90\%} detection accuracy even after attacks.

Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。