Temporal-Sequential Learning With a Brain-Inspired Spiking Neural Network and Its Application to Musical Memory

Sequence learning is an inherent cognitive function of the brain. However, how to represent and memorize sequential information are still fundamental issues in existed models that have not been solved well. To overcome these problems, this paper introduces a spiking neural network inspired by psychological and neurobiological findings. The proposed model have three characteristics: 1) The individual building block of the simulated areas is a neural functional minicolumn which is composed of biologically plausible neurons. 2) Both excitatory and inhibitory connections between neurons are included. The connections between neurons in the same layer are inhibitory and the ones between neurons from different layers are excitatory. These connections are modulated dynamically using the Spike Timing-Dependent Plasticity(STDP) learning rule. 3) The mechanism of temporal sequential patterns is introduced, which is important in sequence learning but ignored by traditional algorithms. To validate our model, we take the music memory to verify. Pitches and rhythms correspond to spacial and temporal patterns respectively. The results have shown that the model can store the input melodies and recall them with high accuracy. Besides, the model can remember the whole melody only given an episode.