Notes On Quantized Convolutional Neural Networks

Existing works:

(1) Speed-up convolutional layers: Low-rank approximation and Tensor decomposition

(2) Reduce storage consumption in fully-connected layers: Parameter compression

Drawback: hard to achieve significant acceleration and compression simultaneously for the whole network.