Multi-objective magnitude-based pruning for latency-aware deep neural network compression

Hong，Wenjing1,2,3; Yang，Peng1; Wang，Yiwen4; Tang，Ke1

10.1007/978-3-030-58112-1_32

2020

0302-9743

1611-3349

LECT NOTES ARTIF INT

12269 LNCS

470-483

Layer-wise magnitude-based pruning is a popular method for Deep Neural Network (DNN) compression. It has the potential to reduce the latency for an inference made by a DNN by pruning connects in the network, which prompts the application of DNNs to tasks with real-time operation requirements, such as self-driving vehicles, video detection and tracking. However, previous methods mainly use the compression rate as a proxy for the latency, without explicitly accounting for latency in the training of the compressed network. This paper presents a new layer-wise magnitude-based pruning method, namely Multi-objective Magnitude-based Latency-Aware Pruning (MMLAP). MMLAP captures latency directly and incorporates a novel multi-objective evolutionary algorithm to optimize both accuracy of a DNN and its latency efficiency when designing compressed networks, i.e., when tuning hyper-parameters of LMP. Empirical studies show the competitiveness of MMLAP compared to well-established LMP methods and show the value of multi-objective optimization in yielding Pareto-optimal compressed networks in terms of accuracy and latency.

Compression Latency-aware Magnitude-based pruning Multi-objective evolutionary algorithm Multi-objective optimization

第一 ; 通讯

英语

EI

20203909228657

Multilayer neural networks ; Pareto principle ; Deep neural networks ; Evolutionary algorithms

Ergonomics and Human Factors Engineering:461.4 ; Optimization Techniques:921.5

2-s2.0-85091285814

Scopus

1.Guangdong Provincial Key Laboratory of Brain-Inspired Intelligent Computation,Department of Computer Science and Engineering,Southern University of Science and Technology,Shenzhen,518055,China
2.Department of Management Science,University of Science and Technology of China,Hefei,230027,China
3.Guangdong-Hong Kong-Macao Greater Bay Area Center for Brain Science and Brain-Inspired Intelligence,Guangzhou,510515,China
4.Department of Electronic and Computer Engineering,Department of Chemical and Biological Engineering,Hong Kong University of Science and Technology,Hong Kong

Hong，Wenjing,Yang，Peng,Wang，Yiwen,et al. Multi-objective magnitude-based pruning for latency-aware deep neural network compression[C],2020:470-483.