Workload automatic mapper for spiking neural network based on precise communication modeling

doi:10.11772/j.issn.1001-9081.2022010078

Journal of Computer Applications ›› 2023, Vol. 43 ›› Issue (3): 827-834.DOI: 10.11772/j.issn.1001-9081.2022010078

• Advanced computing • Previous Articles

Workload automatic mapper for spiking neural network based on precise communication modeling

Xia HUA¹, Zhenghao ZHU¹, Cong XU¹, Xihuang ZHANG¹(), Zhilei CHAI¹^,², Wenjie CHEN³

^1.School of Artificial Intelligence and Computer Science，Jiangnan University，Wuxi Jiangsu 214122，China
^2.Jiangsu Provincial Engineering Laboratory of Pattern Recognition and Computational Intelligence （Jiangnan University），Wuxi Jiangsu 214122，China
^3.MoE Engineering Research Center for Software/Hardware Co-design Technology and Application （East China Normal University），Shanghai 200062，China

Received:2022-01-20 Revised:2022-04-06 Accepted:2022-04-14 Online:2022-05-05 Published:2023-03-10
Contact: Xihuang ZHANG
About author:HUA Xia， born in 1997， M. S. candidate. His research interests include neuromorphic computing.
ZHU Zhenghao， born in 1996， M. S. candidate. His research interests include neuromorphic computing.
XU Cong， born in 1997， M. S. candidate. His research interests include neuromorphic computing.
CHAI Zhilei， born in 1975， Ph. D.， professor. His research interests include computer architecture.
CHEN Wenjie， born in 1977， Ph. D.， associate professor. His research interests include embedded system.
Supported by:
National Natural Science Foundation of China(61972180)

基于精准通信建模的脉冲神经网络工作负载自动映射器

华夏¹, 朱铮皓¹, 徐聪¹, 张曦煌¹(), 柴志雷¹^,², 陈闻杰³

^1.江南大学人工智能与计算机学院, 江苏无锡 214122
^2.江苏省模式识别与计算智能工程实验室(江南大学), 江苏无锡 214122
^3.软硬件协同设计技术与应用教育部工程研究中心(华东师范大学), 上海 200062

通讯作者: 张曦煌
作者简介:华夏（1997—），男，江苏无锡人，硕士研究生，CCF会员，主要研究方向：类脑计算
朱铮皓（1996—），男，江苏无锡人，硕士研究生，主要研究方向：类脑计算
徐聪（1997—），男，浙江湖州人，硕士研究生，CCF会员，主要研究方向：类脑计算
柴志雷（1975—），男，山西新绛人，教授，博士，CCF会员，主要研究方向：计算机体系结构
陈闻杰（1977—），男，浙江金华人，副教授，博士，主要研究方向：嵌入式系统。
基金资助:
国家自然科学基金资助项目(61972180)

Abstract

Abstract:

Running a large-scale Spiking Neural Network （SNN） on a distributed computing platform is one of the basic means to improve the level of brain-like computing intelligence. The difficulty lies in how to deploy the SNN to the corresponding number of computing nodes in order to make the overall system run with the best energy efficiency. To solve this problem， on the basis of NEural Simulation Tool-based （NEST-based） Workload Automatic Mapper for SNN （SWAM） proposed by others before， a workload automatic mapper for SNN， named SWAM2， based on precise communication modeling was proposed. In SWAM2， based on the NEST simulator， the communication part of the SNN workload was further accurately modeled； the quantization method of the parameters in the workload model was improved； the maximum network scale prediction method was designed. Experimental results on typical cases of SNN show that， the average prediction errors of SWAM2 were reduced by about 12.62 and 5.15 percentage points respectively compared with those of SWAM in workload communication and computing time prediction. When predicting the optimal mapping of the workload， the average accuracy of SWAM2 reached 97.55%， which was 13.13 percentage points higher than that of SWAM. SWAM2 avoids the process of manual trial and error by automatically predicting the optimal deployment/mapping of SNN workload on computing platform.

Key words: Spiking Neural Network (SNN), workload mapping, distributed computing platform, NEural Simulation Tool (NEST) simulator, calculation energy efficiency

摘要：

在分布式计算平台上运行大规模的脉冲神经网络（SNN）是提升类脑计算智能水平的基本手段之一，它的难点在于如何将SNN部署到对应数量的计算节点上，使整体系统的运行能效最佳。针对以上问题，在基于NEST的SNN工作负载自动映射器（SWAM）的基础上，提出一种基于精准通信建模的SNN工作负载自动映射器（SWAM2）。在SWAM2中，基于NEST仿真器对SNN工作负载的通信部分进行精准建模，并改进工作负载模型中参数的量化方法，设计了最大网络规模预测方法。在SNN典型案例上的实验结果表明，在工作负载通信以及计算时间的预测中，SWAM2的平均预测误差比SWAM分别降低12.62和5.15个百分点；在对工作负载最佳映射的预测中，SWAM2的平均准确率为97.55%，比SWAM高13.13个百分点。SWAM2通过自动预测SNN工作负载在计算平台上的最佳部署/映射，避免了手动反复实验的过程。

关键词: 脉冲神经网络, 工作负载映射, 分布式计算平台, NEST仿真器, 计算能效

CLC Number:

TP302

Xia HUA, Zhenghao ZHU, Cong XU, Xihuang ZHANG, Zhilei CHAI, Wenjie CHEN. Workload automatic mapper for spiking neural network based on precise communication modeling[J]. Journal of Computer Applications, 2023, 43(3): 827-834.

华夏, 朱铮皓, 徐聪, 张曦煌, 柴志雷, 陈闻杰. 基于精准通信建模的脉冲神经网络工作负载自动映射器[J]. 《计算机应用》唯一官方网站, 2023, 43(3): 827-834.

Figures/Tables 14

Fig. 1 NEST operation mechanism

Fig. 2 SWAM flow

Tab. 1 Classification of workload model parameters

负载模型	时间参数	网络参数
$T n$	$t n_u n r e f (i) 、 t n_r e f (i) 、 t s p i k e (i)$	$N s p i k e (i)$
$T s$	$t s (i)$	$N s p i k e (j p r e) 、 A v g o u t (j p r e) (z)$
$t i m e M P I$	$t a 、 t b$	$s e n d b u f f e r$

Tab. 1 Classification of workload model parameters

负载模型	时间参数	网络参数
$T n$	$t n_u n r e f (i) 、 t n_r e f (i) 、 t s p i k e (i)$	$N s p i k e (i)$
$T s$	$t s (i)$	$N s p i k e (j p r e) 、 A v g o u t (j p r e) (z)$
$t i m e M P I$	$t a 、 t b$	$s e n d b u f f e r$

Fig. 3 Design of spike gathering stage in Single_Mode

Fig. 4 SWAM2 flow

Fig. 5 PYNQ brain-like computing platform

Tab. 2 Comparison of sendbuffer in Example1

预测方式	不同进程数下的send_buffer
预测方式	6	10	40	70	100
SWAM2	5 512	2 280	620	508	220
SWAM	4 804	1 868	564	252	148
实测	5 648	2 484	680	524	300

Tab. 3 Comparison of sendbuffer in Example2

预测方式	不同进程数下的send_buffer
预测方式	4	10	40	70	100
SWAM2	9 550	3 818	958	555	391
SWAM	8 320	3 080	960	480	368
实测	9 632	3 816	1 200	688	448

Tab. 4 Comparison of predicted accuracy

预测方式	案例	$s e n d b u f f e r$	$N s p i k e$	$t i m e M P I$	$t i m e c a l c u l a t e$
SWAM2	Example1	94.85	97.54	97.39	98.41
SWAM2	Example2	96.75	96.47	97.41	97.64
SWAM	Example1	79.24	92.42	85.24	92.12
SWAM	Example2	83.67	93.15	84.32	93.63

Tab. 4 Comparison of predicted accuracy

预测方式	案例	$s e n d b u f f e r$	$N s p i k e$	$t i m e M P I$	$t i m e c a l c u l a t e$
SWAM2	Example1	94.85	97.54	97.39	98.41
SWAM2	Example2	96.75	96.47	97.41	97.64
SWAM	Example1	79.24	92.42	85.24	92.12
SWAM	Example2	83.67	93.15	84.32	93.63

Fig. 6 Comparison of SWAM2 and SWAM with measured data

Fig. 7 Comparison of SWAM2 and SWAM predicted simulation time with measured data

Tab. 5 Comparison of best mapping results

案例	计算平台	SWAM2	SWAM	实测
Example1	ARM	32	36	30
Example1	ARM+FPGA	20	24	20
Example2	ARM	49	53	48
Example2	ARM+FPGA	24	28	24

Tab. 6 Comparison between platform’s NM and measured data

计算平台	不同投入节点数下的N_M
计算平台	10	30	50	70	100
ARM_P	2 719	2 633	2 553	2 478	2 372
ARM_M	2 700	2 588	2 475	2 395	2 250
ARM+FPGA_P	2 175	2 107	2 043	1 982	1 898
ARM+FPGA_M	2 138	2 063	2 025	1 929	1 800

Tab. 7 SWAM2 memory consumption

案例	V_static	V_collect	总内存消耗
Example1	4.6	176.4	181.1
Example2	31.2	9 647.1	9 678.3

References 28

1	MAASS W. Networks of spiking neurons： the third generation of neural network models［J］. Neural Networks， 1997， 10（9）： 1659-1671. 10.1016/s0893-6080(97)00011-7
2	KNIGHT J C， NOWOTNY T. Larger GPU-accelerated brain simulations with procedural connectivity［J］. Nature Computational Science， 2021， 1（2）： 136-142. 10.1038/s43588-020-00022-7
3	WEIDEL P， DUARTE R， MORRISON A. Unsupervised learning and clustered connectivity enhance reinforcement learning in spiking neural networks［J］. Frontiers in Computational Neuroscience， 2021， 15： No.543872. 10.3389/fncom.2021.543872
4	GEWALTIG M O， DIESMANN M. NEST （NRural Simulation Tool）［J］. Scholarpedia， 2007， 2（4）： No.1430. 10.4249/scholarpedia.1430
5	STIMBERG M， BRETTE R， GOODMAN D F M. Brian 2， an intuitive and efficient neural simulator［J］. eLife， 2019， 8： No.e47314. 10.7554/elife.47314
6	STIMBERG M， GOODMAN D F M， NOWOTNY T. Brian2GeNN： accelerating spiking neural network simulations with graphics hardware［J］. Scientific Reports， 2020， 10（1）： No.410. 10.1038/s41598-019-54957-7
7	CHOU T S， KASHYAP H J， XING J W， et al. CARLsim 4： an open source library for large scale， biologically detailed spiking neural network simulation using heterogeneous clusters［C］// Proceedings of the 2018 International Joint Conference on Neural Networks. Piscataway： IEEE， 2018： 1-8. 10.1109/ijcnn.2018.8489326
8	YAKOPCIC C， RAHMAN N， ATAHARY T， et al. Solving constraint satisfaction problems using the Loihi spiking neuromorphic processor［C］// Proceedings of the 2020 Design， Automation and Test in Europe Conference and Exhibition. Piscataway： IEEE， 2020： 1079-1084. 10.23919/date48585.2020.9116227
9	DENG L， WANG G R， LI G Q， et al. Tianjic： a unified and scalable chip bridging spike-based and continuous neural computation［J］. IEEE Journal of Solid-State Circuits， 2020， 55（8）： 2228-2246. 10.1109/jssc.2020.2970709
10	FURBER S B， GALLUPPI F， TEMPLE S， et al. The SpiNNaker project［J］. Proceedings of the IEEE， 2014， 102（5）： 652-665. 10.1109/jproc.2014.2304638
11	POTJANS T C， DIESMANN M. The cell-type specific cortical microcircuit： relating structure and activity in a full-scale spiking network model［J］. Cerebral Cortex， 2014， 24（3）： 785-806. 10.1093/cercor/bhs358
12	刘俊秀，黄星月，罗玉玲，等. 脉冲神经网络硬件互连系统的动态优先级仲裁策略［J］. 电子学报， 2018， 46（8）：1898-1905. 10.3969/j.issn.0372-2112.2018.08.014
	LIU J X， HUANG X Y， LUO Y L， et al. Dynamic priority arbitration strategy for interconnections of hardware spiking neural networks［J］. Acta Electronica Sinica， 2018， 46（8）： 1898-1905. 10.3969/j.issn.0372-2112.2018.08.014
13	BALAJI A， DAS A， WU Y F， et al. Mapping spiking neural networks to neuromorphic hardware［J］. IEEE Transactions on Very Large Scale Integration （VLSI） Systems， 2020， 28（1）： 76-86. 10.1109/tvlsi.2019.2951493
14	刘家华，陈靖宇. 多核并行脉冲神经网络模拟器的设计［J］. 计算机工程与应用， 2020， 56（22）：244-250. 10.3778/j.issn.1002-8331.1908-0375
	LIU J H， CHEN J Y. Design of multi-core parallel spiking neural network simulator［J］. Computer Engineering and Applications， 2020， 56（22）： 244-250. 10.3778/j.issn.1002-8331.1908-0375
15	QU P， ZHANG Y H， FEI X， et al. High performance simulation of spiking neural network on GPGPUs［J］. IEEE Transactions on Parallel and Distributed Systems， 2020， 31（11）： 2510-2523. 10.1109/tpds.2020.2994123
16	TITIRSHA T， SONG S， BALAJI A， et al. On the role of system software in energy management of neuromorphic computing［C］// Proceedings of the 18th ACM International Conference on Computing Frontiers. New York： ACM， 2021： 124-132. 10.1145/3457388.3458664
17	SONG S H， VARSHIKA M L， DAS A， et al. A design flow for mapping spiking neural networks to many-core neuromorphic hardware［C］// Proceedings of the 2021 IEEE/ACM International Conference on Computer Aided Design. Piscataway： IEEE， 2021： 1-9. 10.1109/iccad51958.2021.9643500
18	KULKARNI S R， PARSA M， MITCHELL J P， et al. Benchmarking the performance of neuromorphic and spiking neural network simulators［J］. Neurocomputing， 2021， 447： 145-160. 10.1016/j.neucom.2021.03.028
19	KUNKEL S， SCHENCK W. The NEST dry-run mode： efficient dynamic analysis of neuronal network simulation code［J］. Frontiers in Neuroinformatics， 2017， 11： No.40. 10.3389/fninf.2017.00040
20	NGUYEN Q A P， ANDELFINGER P， TAN W J， et al. Transitioning spiking neural network simulators to heterogeneous hardware［J］. ACM Transactions on Modeling and Computer Simulation， 2021， 31（2）： No.9. 10.1145/3422389
21	郁龚健，张鲁飞，李佩琦，等. SWAM： SNN工作负载自动映射器［J］. 计算机科学与探索， 2021， 15（9）：1641-1657.
	YU G J， ZHANG L F， LI P Q， et al. SWAM： workload automatic mapper for SNN［J］. Journal of Frontiers of Computer Science and Technology， 2021， 15（9）： 1641-1657.
22	BALAJI P， BLAND W， GROPP W， et al. MPICH user’s guide version 3.1［EB/OL］. （2014-02-20）［2022-03-29］..
23	THAKUR R， RABENSEIFNER R， GROPP W. Optimization of collective communication operations in MPICH［J］. The International Journal of High Performance Computing Applications， 2005， 19（1）： 49-66. 10.1177/1094342005051521
24	CROCKETT L， NORTHCOTE D， RAMSAY C， et al. Exploring Zynq^® MPSoC： with PYNQ and Machine Learning Applications［M］. Glasgow： Strathclyde Academic Media， 2019：525-562.
25	张新伟，李康，郁龚健，等. 基于ZYNQ集群的神经形态计算加速研究与实现［J］. 计算机工程与应用， 2020， 56（21）：65-71. 10.3778/j.issn.1002-8331.1912-0339
	ZHANG X W， LI K， YU G J， et al. Research and implementation of accelerating neuromorphic computing based on ZYNQ cluster［J］. Computer Engineering and Applications， 2020， 56（21）： 65-71. 10.3778/j.issn.1002-8331.1912-0339
26	李佩琦，郁龚健，华夏，等. PEST：由PYNQ集群实现的高能效NEST类脑仿真器［J］. 计算机科学与探索， 2021， 15（11）：2127-2141. 10.3778/j.issn.1673-9418.2011047
	LI P Q， YU G J， HUA X， et al. PEST： energy-efficient NEST brain-like simulator implemented by PYNQ cluster［J］. Journal of Frontiers of Computer Science and Technology， 2021， 15（11）： 2127-2141. 10.3778/j.issn.1673-9418.2011047
27	MORRISON A， AERTSEN A， DIESMANN M. Spike-timing-dependent plasticity in balanced random networks［J］. Neural Computation， 2007， 19（6）： 1437-1467. 10.1162/neco.2007.19.6.1437
28	KHERADPISHEH S R， GANJTABESH M， THORPE S J， et al. STDP-based spiking deep convolutional neural networks for object recognition［J］. Neural Networks， 2018， 99： 56-67. 10.1016/j.neunet.2017.12.005

[1]	WANG Xiuqing, HOU Zengguang, PAN Shiying, TAN Min, WANG Yongji, ZENG Hui. Corridor scene recognition for mobile robots based on multi-sonar-sensor information and NeuCube [J]. Journal of Computer Applications, 2015, 35(10): 2833-2837.
[2]	. Target extraction in infrared image based on spiking neural networks [J]. Journal of Computer Applications, 2010, 30(12): 3327-3330.

Workload automatic mapper for spiking neural network based on precise communication modeling

基于精准通信建模的脉冲神经网络工作负载自动映射器

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 14

References 28

Related Articles 2

Recommended Articles

Metrics