0f4d1fc085b7504c140e66bb26ed8842-Paper-Conference.pdf
–Neural Information Processing Systems
Empirical works show that for ReLU neural networks (NNs) with small initialization, input weights of hidden neurons (the input weight of a hidden neuron consists oftheweight fromitsinputlayertothehidden neuron anditsbiasterm) condense ontoisolated orientations.
Neural Information Processing Systems
Feb-7-2026, 11:37:50 GMT
- Technology: