Deep Unitary Convolutional Neural Networks

Hao-Yuan Chang, Kang L. Wang

Deep neural networks can suffer from the exploding and vanishing activation problem, in which the networks fail to train properly because the neural signals either amplify or atten…