Reproduce “MTL-UE: Learning to Learn Nothing for Multi-Task Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yi Yu, Song Xia, Siyuan Yang, Chenqi Kong, Wenhan Yang, Shijian Lu, Yap-Peng Tan, Alex C. Kot
A framework called MTL-UE enables protection of multi-task learning datasets through generated unlearnable examples, combining task-specific embeddings with intra/inter-task regularization to prevent unauthorized model training while maintaining flexibility for partial task protection across different architectures and datasets.
Key insights
- Task label priors and class-wise feature embeddings help reduce the perturbation search space
- Balancing intra-class variance and inter-class separation is crucial for effective multi-task protection
- Framework can support partial protection by making specific tasks unlearnable while preserving others
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.