Dynamic threshold integrate and fire neuron model for low latency spiking neural networks.

•We theoretically analyze conventional ANN-to-SNN conversion that maps ReLU to IF.•We propose a new bio-plausible spiking neuron model called DTIF where the threshold is inversely related to the neuron input at each time-step.•We implement a new conversion mapping ReLU to DTIF and show that the DTIF model outperforms the IF model for the latency performance.