Spiking Reservoir Neural Network for Time Series Classification

Brain-inspired reservoir computing methods attracted a great attention due to their reduced computation complexity by using fixed internal synaptic strengths. We consider a reservoir neural network of spiking neurons supervised trained to use it as a feature extraction layer for solving time-series classification tasks. The original time series input is first encoded into the multiple spike streams to feed it into the spiking reservoir layer. We conducted experiments on ambulatory ECG recordings. The considered approach demonstrates competitive accuracy and robustness over other existing methods.