The integration of new information during sleep reshapes cortical representations that support categorical knowledge. Auto-associative attractor network theories predict that reciprocal excitatory connections help form stable categorical attractors, but direct evidence is missing. We tested this using ten weeks of enriched experience (ENR) in mice as a model for knowledge accumulation and recorded single-unit activity across hippocampus and neocortex. ENR induced significant remodeling in high- but not low-level neocortex, with a shift from unidirectional to bidirectional excitatory-excitatory connections, suggestive of increased \"cell assemblies\". This was accompanied by increased inhibitory-to-excitatory connections and sparser, more orthogonal population activity during awake rest and slow-wave sleep, particularly in deep layers. Thus, ENR reorganizes cortical circuits into a symmetric, inhibition-balanced network that improves coding efficiency, supporting long-standing attractor network predictions.