Summary of Privacy-preserving Student Learning with Differentially Private Data-free Distillation, by Bochao Liu and Jianghu Lu and Pengju Wang and Junjie Zhang and Dan Zeng and Zhenxing Qian and Shiming Ge
Privacy-Preserving Student Learning with Differentially Private Data-Free Distillation
by Bochao Liu, Jianghu Lu, Pengju Wang, Junjie Zhang, Dan Zeng, Zhenxing Qian, Shiming Ge
First submitted to arxiv on: 19 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper presents an innovative teacher-student learning approach to train deep learning models that balance high inference accuracy with data privacy protection. The method utilizes differentially private data-free distillation, where synthetic data is generated to mimic a well-trained teacher model. A generator is pre-trained in a data-free manner using the teacher as a fixed discriminator, allowing for massive synthetic data generation without compromising data privacy. The approach also incorporates selective randomized response label differential privacy algorithm to protect private labels. Experimental results demonstrate the effectiveness of this unified framework in preserving both data and label privacy. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a way to train deep learning models that are super accurate but also keep our personal information safe. They do this by making fake data that’s similar to real data, but not actually the same. This fake data is used to train a new model that can mimic the behavior of the original model, but without using any actual private data. The paper also includes ways to protect the labels or answers we get from using these models. Overall, this approach helps keep our personal information safe while still allowing us to use powerful machine learning models. |
Keywords
» Artificial intelligence » Deep learning » Distillation » Inference » Machine learning » Synthetic data » Teacher model
