Manifold MixUp

• mixes feature values generated from intermediate neural network layers.
• feedforward up to k layer of the network where the output feature maps are mixed
• The mixed feature maps are given input to the next layer and forward propagated up to the last layer.
• After the forward propagation, backward propagation is performed in the standard way with updated labels