ddobbelaere

It might be desirable to monitor/detect vanishing gradients during training. Note that I of course mean "stochastic gradient" here, as estimated by the training samples used in the current epoch...