Paper published on Arxiv

I have published a paper version of my patent on Arxiv, titled “Iterative Training: Finding Binary Weight Deep Neural Networks with Layer Binarization”.

It is a way to train binary weight neural networks with better accuracy at an expense of training time. It details an implementation of the patent. The source code is open-source.

Links:

• Arxiv: https://arxiv.org/abs/2111.07046
• GitHub: https://github.com/rakutentech/iterative_training
• Patent: https://patents.google.com/patent/JP6795721B1/en