notesum.ai

Published at November 11

Evolving Efficient Genetic Encoding for Deep Spiking Neural Networks

cs.NE

cs.AI

Released Date: November 11, 2024

Authors: Wenxuan Pan¹, Feifei Zhao¹, Bing Han¹, Haibo Tong¹, Yi Zeng¹

Aff.: ¹Unknown

Arxiv: http://arxiv.org/abs/2411.06792v1

Refer to caption

Methods	Architecture	Param (M)	Timestep	CIFAR10(%)	CIFAR100(%)
STBP [6]	CIFARNet	17.54	12	89.83	-
STBP NeuNorm [38]	CIFARNet	17.54	12	90.53	-
TSSL-BP [36]	CIFARNet	17.54	5	91.41	-
Our Method	CIFARNet	3.72/3.81	4	93.47	72.97
Our Method	CIFARNet	7.44/7.53	4	95.79	75.05
Hybrid training [24]	VGG-11	9.27	125	92.22	67.87
ANN2SNN [37]	VGG-16	33.6	32	95.42	77.34
Our Method	VGG-16	2.51/2.70	4	94.47	75.02
Our Method	VGG-16	6.89/7.26	4	95.57	76.94
Diet-SNN [34]	ResNet-20	0.27	10/5	92.54	64.07
STBP-tdBN [33]	ResNet-19	12.63	4	92.92	70.86
TET [25]	ResNet-19	12.63	4	94.44	74.47
Spikformer [29]	Spikformer-4-256	4.15	4	93.94	75.96
Spikformer [29]	Spikformer-4-384	9.32	4	95.19	77.86
GAC-SNN [27]	MS-ResNet-18	12.63/12.67	4	96.24	79.83
Our Method	ResNet-18	3.66/3.68	4	95.60	77.32
	ResNet-18	8.72/8.75	4	96.26	78.91
	ResNet-19	4.35/4.38	4	95.94	77.56
	ResNet-19	8.41/8.44	4	96.80	78.86