Convolutional Neural Networks (CNNs)
WaveNet basic idea is the dilated causal convolution. Instead of squashing all the context through one single layer which makes information extremely lossy, we pass the context through multiple layers of dilated convolutions, where each layer takes a larger context into account. By stacking multiple layers of dilated convolutions with exponentially increasing dilation factors, the receptive field is able to cover very large contexts with relatively few layers.