BiasedPUlearnig: Positive and Unlabeled Data with a Selection Bias (ICLR 2019)

• Paper link: https://openreview.net/pdf?id=rJzLciCqKm
• Author's code repo: https://github.com/MasaKat0/PUlearning. Note that it is implemented by Chainer.

I implemented the nnPUSB: Model with neural Networks by PyTorch and got a better performance.

Comparison of average results obtained by running experiment on the mnist dataset 10 times.

You can refer to the code of test_mnist.py to train your own model.