1 paper · 1 filter
Alexander Shekhovtsov, Viktor Yanush, Boris Flach
In neural networks with binary activations and or binary weights the training by gradient descent is complicated as the model has piecewise constant response. We consider stochasti…