This paper proposes new method of adaptive threshold adjusting during runtime and neuron pruning for Spiking Neural Networks(SNNs). The proposed adaptive threshold scheme reduces data loss in overly active neurons and decrease conversion accuracy drop...
This paper proposes new method of adaptive threshold adjusting during runtime and neuron pruning for Spiking Neural Networks(SNNs). The proposed adaptive threshold scheme reduces data loss in overly active neurons and decrease conversion accuracy drop by 14.6% and synaptic operations by 5.6%. Additionally, implementing proposed neuron pruning removes inactive neurons, further increasing synaptic operation reduction to 11.6%.