Modelstamp：机器学习工件与运行时环境状态的反序列化前验证
原标题：Modelstamp: Pre-Deserialization Verification of Machine-Learning Artifacts and Runtime Environment State
AI 摘要
arXiv 论文提出 Modelstamp，一种轻量级 Python 持久化库，用于在反序列化前验证机器学习工件的完整性和运行时环境状态。它通过附带 JSON 清单记录 SHA-256 摘要、运行时元数据和包版本，并支持 HMAC 认证。实验表明，在 14 个环境漂移和 8 个信任边界场景中表现符合预期，验证时间从 10 MiB 的 0.032 秒增至 1 GiB 的 3.334 秒，吞吐量约 307-312 MiB/s。该工具作为反序列化前的补充验证控制，而非替代依赖管理系统或恶意模型检测。
正文节选
[Page 1] Modelstamp: Pre-Deserialization Verification of Machine-Learning Artifacts and Runtime Environment State Anagha Dhekne Independent Researcher Abstract—Persisted machine-learning models can remain byte-identical while the software environments in which they are loaded evolve, creating a verification problem that artifact integrity checks alone cannot expose. This paper pre