Learning regularities in noise engages both neural predictive activity and representational changes.

The ability to extract structured sensory patterns from a noisy environment is fundamental to cognition, yet how the brain learns complex regularities remains unclear. Using magnetoencephalography during a visuomotor task, we tracked the neural dynamics as humans learned non-adjacent temporal dependencies embedded in noise. We reveal that learning is supported by two temporally dissociable mechanisms. Neural predictive activity emerged rapidly, with stimulus-specific patterns appearing before stimulus onset and preceding measurable behavioral improvements. This is followed by a slower build-up of representational change, characterized by an increased neural pattern similarity between statistically dependent, non-adjacent elements. Both processes are supported by a distributed consortium of networks, with the sensorimotor and dorsal attentional networks playing a central role. These findings suggest that both neural predictive activity and representational changes contribute to learning regularities, revealing a temporal hierarchy in which neural predictive activity precedes behavioral improvement and is followed by neural representational changes, possibly facilitating the gradual consolidation of knowledge into stable neural representations.