Neural sampling from cognitive maps enables goal-directed imagination and planning
Nature Machine Intelligence, Published online: 21 July 2026; doi:10.1038/s42256-026-01254-4 Lin et al. introduce a brain-inspired generative model that provides two key features of intelligence: planning and problem-solving. It uses cognitive maps, stochastic computing and compositional coding, and requires only local synaptic plasticity.
Neural sampling from cognitive maps enables goal-directed imagination and planning, according to recent research. Our brains use cognitive maps, stochastic computing, and compositional coding to achieve this capability, consuming only 20 W of energy for online learning and instant adjustment to changing contingencies. Cognitive maps, which encode relational information between states and actions, are essential tools in planning.
Experimental data suggests that they can also be used for problem-solving in non-spatial domains. The study integrates cognitive maps, sampling, and compositional computing into a transparent neural network model suitable for implementation by energy-efficient neuromorphic hardware. The model requires only self-supervised local synaptic plasticity and learns autonomously from own experiences.
Testing the model's functional capabilities in various application domains—reinforcement learning, reinforcement learning in non-spatial domains, and compositional computing—demonstrated its adaptability to changes in goals and contingencies. The model's simple and transparent neural networks learn through online methods, mirroring the brain's efficient use of cognitive maps for goal-directed planning and problem-solving.
Written by urgent.news from Nature Machine Intelligence's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
