High-accuracy deep ANN-to-SNN conversion using quantization-aware training framework and calcium-gated bipolar leaky integrate and fire neuron

• Published in: Frontiers in neuroscience Vol. 17; p. 1141701
• Main Authors: Gao, Haoran, He, Junxian, Wang, Haibing, Wang, Tengxiao, Zhong, Zhengqing, Yu, Jianyi, Wang, Ying, Tian, Min, Shi, Cong
• Format: Journal Article
• Language: English
• Published: Switzerland Frontiers Research Foundation 08.03.2023; Frontiers Media S.A
• Subjects: ANN-to-SNN conversion; Approximation; Bias; deep SNNs; Firing pattern; Latency; Neural networks; neuromorphic computing; Neurons; Neuroscience; quantization-aware training; spiking neural network; ANN-to-SNN conversion; deep SNNs; spiking neural network; quantization-aware training; neuromorphic computing
• ISSN: 1662-4548; 1662-453X; 1662-453X
• DOI: 10.3389/fnins.2023.1141701

Spiking neural networks (SNNs) have attracted intensive attention due to the efficient event-driven computing paradigm. Among SNN training methods, the ANN-to-SNN conversion is usually regarded to achieve state-of-the-art recognition accuracies. However, many existing ANN-to-SNN techniques impose lengthy post-conversion steps like threshold balancing and weight renormalization, to compensate for the inherent behavioral discrepancy between artificial and spiking neurons. In addition, they require a long temporal window to encode and process as many spikes as possible to better approximate the real-valued ANN neurons, leading to a high inference latency. To overcome these challenges, we propose a calcium-gated bipolar leaky integrate and fire (Ca-LIF) spiking neuron model to better approximate the functions of the ReLU neurons widely adopted in ANNs. We also propose a quantization-aware training (QAT)-based framework leveraging an off-the-shelf QAT toolkit for easy ANN-to-SNN conversion, which directly exports the learned ANN weights to SNNs requiring no post-conversion processing. We benchmarked our method on typical deep network structures with varying time-step lengths from 8 to 128. Compared to other research, our converted SNNs reported competitively high-accuracy performance, while enjoying relatively short inference time steps.