Design and Analysis of a Neural Network Inference Engine Based on Adaptive Weight Compression

摘要

Neural networks generally require significant memory capacity/bandwidth to store/access a large number of synaptic weights. This paper presents design of an energy-efficient neural network inference engine based on adaptive weight compression using a JPEG image encoding algorithm. To maximize compression ratio with minimum accuracy loss, the quality factor of the JPEG encoder is adaptively controlled depending on the accuracy impact of each block. With 1% accuracy loss, the proposed approach achieves$63.4{imes }$compression for multilayer perceptron (MLP) and$31.3 {imes }$for LeNet-5 with the MNIST dataset, and$15.3 {imes }$for AlexNet and$10.2 {imes }$for ResNet-50 with ImageNet. The reduced memory requirement leads to higher throughput and lower energy for neural network inference ($3 {imes }$effective memory bandwidth and$22 {imes }$lower system energy for MLP).