LIO: Learning from Indirect Observations

A package for weakly supervised learning research based on PyTorch

Installation

pip install lio

or

git clone https://github.com/YivanZhang/lio.git
pip install -e .

Most of the modules are designed as small (higher-order) functions.
Feel free to copy-paste only what you need for your existing workflow to reduce dependencies.

References

• Approximating Instance-Dependent Noise via Instance-Confidence Embedding
  Yivan Zhang and Masashi Sugiyama
  [arXiv] [code]

• Learning Noise Transition Matrix from Only Noisy Labels via Total Variation Regularization
  Yivan Zhang, Gang Niu, and Masashi Sugiyama
  [arXiv] [ICML'21] [poster] [code]

• Learning from Aggregate Observations
  Yivan Zhang, Nontawat Charoenphakdee, Zhenguo Wu, and Masashi Sugiyama
  [arXiv] [NeurIPS'20] [poster]

• Learning from Indirect Observations
  Yivan Zhang, Nontawat Charoenphakdee, and Masashi Sugiyama
  [arXiv]