检测模型是否通过知识蒸馏复制，无需访问教师模型权重。
Knowledge Distillation Detection for Open-weights Models
- 仅用学生模型权重和教师API，通过无数据输入合成与统计分析检测蒸馏痕迹。
- 在CIFAR-10、ImageNet和文生图任务上，检测准确率比最强基线提升59.6%~71.2%。
- 适用于分类与生成模型，适合关注模型版权与安全的开发者使用。
我们提出知识蒸馏检测任务，旨在判断一个学生模型是否由给定教师模型蒸馏而来，且在仅可访问学生模型权重和教师API的实际场景下进行。该问题源于对模型来源追踪及未经授权蒸馏复制的日益担忧。为此，我们提出一种模型无关框架，结合无数据输入合成与统计得分计算实现蒸馏检测。该方法适用于分类与生成模型。在图像分类与文本到图像生成多种架构上的实验表明，我们的方法在CIFAR-10上检测准确率较最强基线提升59.6%，在ImageNet上提升71.2%，在文生图任务中提升20.0%。代码已公开于https://github.com/shqii1j/distillation_detection。
原文摘要 · Abstract (English)
We propose the task of knowledge distillation detection, which aims to determine whether a student model has been distilled from a given teacher, under a practical setting where only the student's weights and the teacher's API are available. This problem is motivated by growing concerns about model provenance and unauthorized replication through distillation. To address this task, we introduce a model-agnostic framework that combines data-free input synthesis and statistical score computation for detecting distillation. Our approach is applicable to both classification and generative models. Experiments on diverse architectures for image classification and text-to-image generation show that our method improves detection accuracy over the strongest baselines by 59.6% on CIFAR-10, 71.2% on ImageNet, and 20.0% for text-to-image generation. The code is available at https://github.com/shqii1j/distillation_detection.
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。