Fetching the paper…

ReLU Neural Networks with Linear Layers are Biased Towards Single- and Multi-Index Models · Around