Quantisation and Pruning for Neural Network Compression and Regularisation

14 Jan 2020  ·  Kimessha Paupamah, Steven James, Richard Klein ·

Deep neural networks are typically too computationally expensive to run in real-time on consumer-grade hardware and low-powered devices. In this paper, we investigate reducing the computational and memory requirements of neural networks through network pruning and quantisation. We examine their efficacy on large networks like AlexNet compared to recent compact architectures: ShuffleNet and MobileNet. Our results show that pruning and quantisation compresses these networks to less than half their original size and improves their efficiency, particularly on MobileNet with a 7x speedup. We also demonstrate that pruning, in addition to reducing the number of parameters in a network, can aid in the correction of overfitting.

PDF Abstract

Results from the Paper

 Ranked #1 on Network Pruning on CIFAR-10 (Inference Time (ms) metric)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Benchmark
Network Pruning CIFAR-10 MobileNet – Quantised Inference Time (ms) 4.74 # 1
Neural Network Compression CIFAR-10 ShuffleNet – Quantised Size (MB) 1.9 # 1
Network Pruning CIFAR-10 AlexNet – Quantised Inference Time (ms) 5.23 # 2
Network Pruning CIFAR-10 ShuffleNet – Quantised Inference Time (ms) 23.15 # 3
Neural Network Compression CIFAR-10 AlexNet – Quantised Size (MB) 54.6 # 3
Neural Network Compression CIFAR-10 MobileNet – Quantised Size (MB) 2.9 # 2