In this university project we aim to extend a pre-existing implementation of a basic CNN [https://github.com/adityagaydhani/CSCI-5561-CNN-implementation-from-scratch] in order for it to be able to handle Dual Numbers. The focus is to try to exploit the Dual Numbers' "Automatic Differentiation" characteristic, which comes handy in neural network as it lets to compute the gradients during the forward propagation and thus avoiding to compute them during the backward pass.
enricollen / cnn-dual-numbers Goto Github PK
View Code? Open in Web Editor NEWFeasibility Study of CNN with Dual Numbers