Spiking Neural Networks Subject to Adversarial Attacks in Spiking Domain

Lin, Xuanwei; Dong, Chen; Liu, Ximeng; Chen, Dong

doi:10.1007/978-3-031-20096-0_34

Cited by 2 publications

(5 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the performance of the converted SNN is improved, the cost of large time delays also comes. (Deng & Gu, 2021) divide conversion into floor and clip error from a new quantization perspective, (Li et al, 2021a) further optimize the conversion error. (Yu et al, 2021) construct the deep SNN with doublethreshold, (Liu et al, 2022) propose temporal separation to further zip gap between ANN and SNN.…”

Section: Related Workmentioning

confidence: 99%

“…Nonetheless, most previous works have been limited to the fixed threshold and ignored utilizing the individual neurons to transmit information effectively. Some studies (Kim et al, 2020;Li et al, 2021a) take threshold variation into consideration while they still use the two-stage or heuristic method for SNNs, which require careful design or extensive search. (Ding et al, 2022) use the biologically dynamic threshold mechanism to do robot obstacle avoidance and continuous control tasks and enhance the host SNN generalization.…”

Section: Use Burst Mechanism and Proposementioning

confidence: 99%

“…There is no closed-form solution to the above problem, Li et al (2021a) use gird search to heuristically find the final solution. A trivial solution is to make…”

Section: Adaptive Threshold Optimization Errormentioning

confidence: 99%

“…In order not to do too many restrictions on origin ANN, we choose to save the topology and use the original weights in the target ANN for universal conversion without retraining ANN. In this way, we compare the state-of-the-art approaches including TSC , RMP-SNN , Opt (Deng & Gu, 2021), Calibration (Li et al, 2021a), Burst (Li & Zeng, 2022).…”

Section: Comparison With the State Of The Artmentioning

confidence: 99%

“…As shown in Table 4, we also achieve comparative performance with SOTA methods and nearly lossless conversion from the target ANN. It is worth noting that though calibration (Li et al, 2021a) also uses the dynamic threshold, our method does not need to search for the best voltage threshold, which could result in less computational cost.…”

Section: Comparison With the State Of The Artmentioning

confidence: 99%

See 4 more Smart Citations

MSAT: Biologically Inspired Multi-Stage Adaptive Threshold for Conversion of Spiking Neural Networks

He¹,

Li²,

Zhao³

et al. 2023

Preprint

View full text Add to dashboard Cite

Spiking Neural Networks (SNNs) can do inference with low power consumption due to their spike sparsity. ANN-SNN conversion is an efficient way to achieve deep SNNs by converting well-trained Artificial Neural Networks (ANNs). However, the existing methods commonly use constant threshold for conversion, which prevents neurons from rapidly delivering spikes to deeper layers and causes high time delay. In addition, the same response for different inputs may result in information loss during the information transmission. Inspired by the biological model mechanism, we propose a multi-stage adaptive threshold (MSAT). Specifically, for each neuron, the dynamic threshold varies with firing history and input properties and is positively correlated with the average membrane potential and negatively correlated with the rate of depolarization. The self-adaptation to membrane potential and input allows a timely adjustment of the threshold to fire spike faster and transmit more information. Moreover, we analyze the Spikes of Inactivated Neurons error which is pervasive in early time steps and propose spike confidence accordingly as a measurement of confidence Xiang He and Yang Li contributed equally to this work.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Use Burst Mechanism and Proposementioning

confidence: 99%

“…There is no closed-form solution to the above problem, Li et al (2021a) use gird search to heuristically find the final solution. A trivial solution is to make…”

Section: Adaptive Threshold Optimization Errormentioning

confidence: 99%