Why the Brain Consolidates: Predictive Forgetting for Optimal Generalisation

TL;DR

This paper proposes that the brain uses predictive forgetting to optimize memory representations for better generalisation, supported by theoretical analysis and simulations across neural and language models.

Contribution

It introduces the concept of predictive forgetting as a mechanism for memory consolidation that enhances generalisation, supported by formal bounds and diverse model simulations.

Findings

Predictive forgetting improves information-theoretic generalisation bounds.

High-capacity networks benefit from iterative, offline refinement of stored traces.

Simulation results align with predictions on neural and language models.

Abstract

Standard accounts of memory consolidation emphasise the stabilisation of stored representations, but struggle to explain representational drift, semanticisation, or the necessity of offline replay. Here we propose that high-capacity neocortical networks optimise stored representations for generalisation by reducing complexity via predictive forgetting, i.e. the selective retention of experienced information that predicts future outcomes or experience. We show that predictive forgetting formally improves information-theoretic generalisation bounds on stored representations. Under high-fidelity encoding constraints, such compression is generally unattainable in a single pass; high-capacity networks therefore benefit from temporally separated, iterative refinement of stored traces without re-accessing sensory input. We demonstrate this capacity dependence with simulations in autoencoder-based neocortical models, biologically plausible predictive coding circuits, and Transformer-based language models, and derive quantitative predictions for consolidation-dependent changes in neural representational geometry. These results identify a computational role for off-line consolidation beyond stabilisation, showing that outcome-conditioned compression optimises the retention-generalisation trade-off.