Accelerating Training of Deep Neural Networks via Sparse Edge Processing

被引:9
作者
Dey, Sourya [1 ]
Shao, Yinan [1 ]
Chugg, Keith M. [1 ]
Beerel, Peter A. [1 ]
机构
[1] Univ Southern Calif, Ming Hsieh Dept Elect Engn, Los Angeles, CA 90089 USA
来源
ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING - ICANN 2017, PT I | 2017年 / 10613卷
关键词
Machine learning; Neural networks; Deep neural networks; Sparsity; Online learning; Training acceleration; Hardware optimizations; Pipelining; Edge processing; Handwriting recognition; IMPLEMENTATION; FPGA;
D O I
10.1007/978-3-319-68600-4_32
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We propose a reconfigurable hardware architecture for deep neural networks (DNNs) capable of online training and inference, which uses algorithmically pre-determined, structured sparsity to significantly lower memory and computational requirements. This novel architecture introduces the notion of edge-processing to provide flexibility and combines junction pipelining and operational parallelization to speed up training. The overall effect is to reduce network complexity by factors up to 30x and training time by up to 35x relative to GPUs, while maintaining high fidelity of inference results. This has the potential to enable extensive parameter searches and development of the largely unexplored theoretical foundation of DNNs. The architecture automatically adapts itself to different network sizes given available hardware resources. As proof of concept, we show results obtained for different bit widths.
引用
收藏
页码:273 / 280
页数:8
相关论文
共 20 条
[1]  
Ahn B, 2014, IEEE IJCNN, P141, DOI 10.1109/IJCNN.2014.6889903
[2]  
[Anonymous], 2013, NIPS
[3]  
[Anonymous], CORR
[4]  
[Anonymous], 2016, ICLR 2016
[5]  
[Anonymous], 1989, NIPS
[6]   DianNao: A Small-Footprint High-Throughput Accelerator for Ubiquitous Machine-Learning [J].
Chen, Tianshi ;
Du, Zidong ;
Sun, Ninghui ;
Wang, Jia ;
Wu, Chengyong ;
Chen, Yunji ;
Temam, Olivier .
ACM SIGPLAN NOTICES, 2014, 49 (04) :269-283
[7]  
Chen WL, 2015, PR MACH LEARN RES, V37, P2285
[8]   DaDianNao: A Machine-Learning Supercomputer [J].
Chen, Yunji ;
Luo, Tao ;
Liu, Shaoli ;
Zhang, Shijin ;
He, Liqiang ;
Wang, Jia ;
Li, Ling ;
Chen, Tianshi ;
Xu, Zhiwei ;
Sun, Ninghui ;
Temam, Olivier .
2014 47TH ANNUAL IEEE/ACM INTERNATIONAL SYMPOSIUM ON MICROARCHITECTURE (MICRO), 2014, :609-622
[9]   Deep, Big, Simple Neural Nets for Handwritten Digit Recognition [J].
Ciresan, Dan Claudiu ;
Meier, Ueli ;
Gambardella, Luca Maria ;
Schmidhuber, Juergen .
NEURAL COMPUTATION, 2010, 22 (12) :3207-3220
[10]  
ELDREDGE JG, 1994, 1994 IEEE INTERNATIONAL CONFERENCE ON NEURAL NETWORKS, VOL 1-7, P2097, DOI 10.1109/ICNN.1994.374538