iceiony/NNPlayground

Summary

Equivalent implementations of XOR in dynamic Neural Network frameworks.
Currently known frameworks are: dynet, chainer, pyTorch.

Details (Network and Training)

IN -> linear(2x2) -> sigmoid -> linear(2x1) -> sigmoid -> OUT
Optimizer: SGD, with learning rate 1
Loss Fun : binary cross entropy
Weigths : initialised from normal distribution (sigma = 1)
Training : 500 itterations
Plots : mean training (with standard error) for 100 training runs