Updating batch update experiment.
Experimenting averaged weights.
Very small full connect relu backprop optimization.
Bug fixing animal machine experiment.
Speeding up depthwise backprop.
Speeding up Pointwise backprop.
Fixing hard to find bug related to calling convention (XMM6/YMM6/ZMM6).
Adding debug tests to convolutional code.
Testing separable convolutions.
readme.txt update.