Compressing Low Precision Deep Neural Networks Using Sparsity-Induced Regularization in Ternary Networks
This work addresses the need for efficient, low-resource neural network inference in hardware implementations, though it is incremental as it builds on existing low-precision and sparsity methods.
The paper tackles the problem of compressing deep neural networks for efficient hardware deployment by introducing a training technique that produces sparse, ternary networks, achieving up to 98% sparsity and 5-11 times size reduction compared to existing binary and ternary models on MNIST and CIFAR10 datasets.
A low precision deep neural network training technique for producing sparse, ternary neural networks is presented. The technique incorporates hard- ware implementation costs during training to achieve significant model compression for inference. Training involves three stages: network training using L2 regularization and a quantization threshold regularizer, quantization pruning, and finally retraining. Resulting networks achieve improved accuracy, reduced memory footprint and reduced computational complexity compared with conventional methods, on MNIST and CIFAR10 datasets. Our networks are up to 98% sparse and 5 & 11 times smaller than equivalent binary and ternary models, translating to significant resource and speed benefits for hardware implementations.