I'm using Encog Machine learning api and it uses a Flat implementation to optimize performance, where it assumes that there is a full connection between two layers.
I want to extend it to support convolutional neural networks, But I've no idea how to update the api to support it.
I've thought about setting the weights of the input that doesn't get into the result of the convolution with zero, but am not sure how this will affect the backpropagation algorithm
