Paper: “Learning both Weights and Connections for Efficient Neural Networks”,NIPS 2015
-
Apply Pruning and retrain on mnist-300-100 model using tensorflow
-
Apply Pruning and retrain on mnist-conv-net model using tensorflow
-
Apply quantization to the weights, use 8 bits to quantize the weights, and found its remarkablely great, only a few loss in accuracy.
quantization: refers: https://petewarden.com/2016/05/03/how-to-quantize-neural-networks-with-tensorflow/