Accurate online training of dynamical spiking neural networks through Forward Propagation Through Time

With recent advances in learning algorithms, recurrent networks of spiking neurons are achieving performance that is competitive with vanilla neural networks. However, these algorithms limited to small simple and modest-length temporal sequences, as they impose high memory requirements, have difficulty training complex neuron models incompatible online learning. Here, we show how the recently developed Forward-Propagation Through Time (FPTT) combined novel liquid time-constant resolves limitations. Applying FPTT such neurons, demonstrate exceedingly long sequences while outperforming current methods approaching or offline on classification tasks. The efficiency robustness enable us directly train a deep performant network for joint object localization recognition, demonstrating ability large-scale dynamic architectures. Memory efficient without compromising accuracy an open challenge neuromorphic computing. Yin colleagues consisting so-called using algorithm called allows state-of-the-art at reduced computational cost compared existing approaches.