Common attentional mechanisms, but distinct computational dynamics, across Statistical and Reward Learning

High-reward locations can be decoded before target onset in an EEG visual-search task after a task-irrelevant, spatially unspecific perturbation, indicating latent reward-based spatial representations.

• Why it matters: It remains unclear whether statistical learning and reward learning guide attention using shared neural and computational mechanisms, or whether they rely on distinct dynamics despite similar outcomes.
• What they did: In Experiment 1, multivariate EEG decoding tested pre-target detectability of reward locations after a visual perturbation; in Experiment 2, EEG decoding and drift-rate modeling compared SL and RL across two sessions.
• The result: SL and RL established comparable neural representations and similarly supported target selection and drift-rate, but RL showed a slower learning-rate than SL, with pupillometry reflecting expectation violations in SL and increased arousal in RL.

AI summary of the abstract · may contain errors · not medical advice · read the paper.