Swing Distillation: A Privacy-Preserving Knowledge Distillation Framework
Li, Junzhuo, Wu, Xinwei, Dong, Weilong, Wu, Shuangzhi, Bian, Chao, Xiong, Deyi
–arXiv.org Artificial Intelligence
Knowledge distillation (KD) has been widely used for model compression and knowledge transfer. Typically, a big teacher model trained on sufficient data transfers knowledge to a small student model. However, despite the success of KD, little effort has been made to study whether KD leaks the training data of the teacher model. In this paper, we experimentally reveal that KD suffers from the risk of privacy leakage. To alleviate this issue, we propose a novel knowledge distillation method, swing distillation, which can effectively protect the private information of the teacher model from flowing to the student model. In our framework, the temperature coefficient is dynamically and adaptively adjusted according to the degree of private information contained in the data, rather than a predefined constant hyperparameter. It assigns different temperatures to tokens according to the likelihood that a token in a position contains private information. In addition, we inject noise into soft targets provided to the student model, in order to avoid unshielded knowledge transfer. Experiments on multiple datasets and tasks demonstrate that the proposed swing distillation can significantly reduce (by over 80% in terms of canary exposure) the risk of privacy leakage in comparison to KD with competitive or better performance. Furthermore, swing distillation is robust against the increasing privacy budget.
arXiv.org Artificial Intelligence
Dec-16-2022
- Country:
- North America
- United States
- Texas > Travis County
- Austin (0.04)
- Pennsylvania > Philadelphia County
- Philadelphia (0.04)
- New York > New York County
- New York City (0.04)
- Minnesota > Hennepin County
- Minneapolis (0.14)
- Michigan > Washtenaw County
- Ann Arbor (0.04)
- California > San Diego County
- San Diego (0.04)
- Texas > Travis County
- Canada > British Columbia
- United States
- Europe
- Spain > Catalonia
- Barcelona Province > Barcelona (0.04)
- Italy > Tuscany
- Florence (0.04)
- France > Provence-Alpes-Côte d'Azur
- Bouches-du-Rhône > Marseille (0.04)
- Belgium > Brussels-Capital Region
- Brussels (0.04)
- Spain > Catalonia
- Asia > China
- Tianjin Province > Tianjin (0.04)
- Hong Kong (0.04)
- Beijing > Beijing (0.04)
- North America
- Genre:
- Research Report (1.00)
- Industry:
- Information Technology > Security & Privacy (1.00)
- Education (1.00)
- Technology:
- Information Technology
- Security & Privacy (1.00)
- Knowledge Management (0.86)
- Data Science > Data Mining
- Big Data (0.41)
- Artificial Intelligence
- Natural Language (1.00)
- Machine Learning (1.00)
- Information Technology