Spiking World Model with Multi-Compartment Neurons for Model-based Reinforcement Learning

Significance Dendritic computation is key to the brain’s ability to integrate information over long timescales. Inspired by this, this study proposes a spiking neural network model that embeds dendritic mechanisms to enhance long-term memory and planning. This biologically grounded approach enables the construction of a spiking world model for model-based reinforcement learning, bridging microscale neural computation with macroscale cognitive function. By applying this model to a challenging decision-making task, we show that it can match the performance of traditional artificial intelligence systems while using brain-like mechanisms. Our work not only improves the capabilities of spiking neural networks but also highlights how complex biological processes, like dendritic integration, can inspire more powerful and efficient computing systems.