Jacob J. W. Bakermans

This paper introduces modular neural state-space models to implement the Bayesian Brain hypothesis, enabling learned and reusable world-model components for flexible cognition.

The paper shows that a hierarchical coordinate system in the human entorhinal cortex supports abstract sequence memories, mirroring spatial memory organization.

The paper shows that hippocampal replay constructs compositional memories of future states by recombining existing primitives, enabling optimal behavior in novel environments without new learning.