deel-ai / deel-lip

Build and train Lipschitz constrained networks: TensorFlow implementation of k-Lipschitz layers
https://deel-ai.github.io/deel-lip/
MIT License
89 stars 10 forks source link

Feat/loss soft hkr #86

Open franckma31 opened 1 year ago

franckma31 commented 1 year ago

Introduce the multiclass loss : MulticlassSoftHKR defined in the paper "On the explainable properties of 1-Lipschitz Neural Networks: An Optimal Transport Perspective" Serrurier et al (Neurips'23) that combine an optimal transport loss with a softmax temperature coefficient (useful when the numer of classes is high like in Imagenet dataset)