The abstract is read at the publisher; the summary is JClub's.
In 107 mice performing the International Brain Laboratory visual decision task, a state-dependent reward prediction error added only a small predictive gain beyond preceding choice and reward, near zero after including the preceding stimulus.
- Why it matters: Choice prediction from behavior may reflect either reward prediction-error learning or simpler use of observable trial history, but it is unclear how much model-derived RPEs add beyond that history.
- What they did: The study analyzed 275,292 held-out trials and compared choice models using an RPE with stimulus-dependent reward expectation, then tested whether adding preceding stimulus, and multi-trial history traces, improved prediction.
- The result: Multi-trial RPE traces with the same retention parameter did not reliably improve prediction, and simulations showed RPE-accumulation learning can be detected only when the agent does not already use trial history directly.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.