Gated Recurrent Spiking Neurons for Solving Partially Observable Markov Decision Processes and Multi-Agent Reinforcement Learning

The authors propose a novel temporal alignment paradigm (TAP) and gated recurrent spiking neurons (GRSN) to address the temporal mismatch issue in spiking reinforcement learning (SRL) algorithms and enhance the memory capacity of spiking neurons, enabling them to effectively solve partially observable Markov decision processes (POMDPs) and multi-agent reinforcement learning (MARL) problems.