Answers are grounded in this paper and cite their sources.
Spiking neural networks (SNNs) offer a promising avenue for energy-efficient computations on neuromorphic hardware, leveraging the unique advantages of spike-based signaling. Despite their potential, SNNs often lag behind artificial neural networks (ANNs) in performance, mainly due to the complexity of effectively translating ANN activation values into the time domain spikes. This challenge arises due to the quantization and unevenness errors that occur when mapping continuous activations to discrete spikes and irregular spike timing. This paper introduces NeuBridge, an innovative ANN-SNN conversion method that utilize adaptive temporal coding to significantly reduce the number of required time steps without compromising accuracy. NeuBridge addresses the errors by employing a decode–encode neuron and adaptive temporal coding, effectively bridging the performance gap between ANNs and SNNs. By establishing an equivalence between quantized ANNs and SNNs and optimizing the temporal coding process, we improve SNN performance with as few as 2 time steps. Empirical evaluations on CIFAR-10 and ImageNet datasets demonstrate that NeuBridge consistently performs better than existing conversion methods in both accuracy and efficiency, achieving high performance within just 3 time steps. NeuBridge code is available at: https://github.com/Intelli-Chip-Lab/NeuBridge .
No takes yet. Share an insight, caveat, or question.
Liu et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: