A Brain-Inspired Spiking Neural Network Model for Decision-Making in Reinforcement Learning: Intelligent Self-Balancing Control

Spiking Neural Networks (SNNs) combined with temporal dynamics and spiking behavior inherent in biological neural systems show better explanations and performance in realtime control. We proposed a self-balancing model equipped with a bio-inspired, reward-modulated SNN that employs Reward-spiking-timing-dependent Plasticity (R-STDP). Self-balancing model allows for bio-inspired learning (synaptic weight modification), in which the data essential for modifying weights is accessible at local level. Local learning mirrors the extraordinary learning processes observed within different sections of the human brain. The proposed method reconstructs the taskoriented state space and reward mechanism to realize a practice of the cross-task model, with an average decision time of 15 milliseconds. This study aims to provide a method to bridge the gap between SNNs and traditional feedback mechanisms in control systems and suggests how bio-inspired approaches can enhance these mechanisms. Cases on the proposing model show the ability of SNNs on control application. SNNs make lowpower embedded real-time control and the practical application of neuromorphic hardware platforms possible in the future.