Characterizing Sparse Connectivity Patterns in Neural Networks
This addresses storage efficiency for neural network practitioners, but is incremental as it builds on existing sparsity techniques.
The paper tackles the problem of reducing parameters in neural networks by using pre-defined sparsity, achieving no loss in accuracy with less than 0.5% connection density in classification layers and less than 5% overall.
We propose a novel way of reducing the number of parameters in the storage-hungry fully connected layers of a neural network by using pre-defined sparsity, where the majority of connections are absent prior to starting training. Our results indicate that convolutional neural networks can operate without any loss of accuracy at less than half percent classification layer connection density, or less than 5 percent overall network connection density. We also investigate the effects of pre-defining the sparsity of networks with only fully connected layers. Based on our sparsifying technique, we introduce the `scatter' metric to characterize the quality of a particular connection pattern. As proof of concept, we show results on CIFAR, MNIST and a new dataset on classifying Morse code symbols, which highlights some interesting trends and limits of sparse connection patterns.