Evidence in favor of abrupt over gradual learning in the differential reinforcement of response duration (DRRD) task

Learning can occur in markedly different ways: in some cases, it unfolds as a gradual process, with behavior improving slowly toward an asymptotic level of performance; in others, it appears as an abrupt process that sharply separates behavior before and after a change point. Under-standing the behavioral and neural processes underlying these distinct acquisition patterns may be critical for elucidating the basic principles of learning. We investigated this question experimentally using naive rats performing a differential reinforcement of response duration (DRRD) task, in which animals were required to remain inside a nosepoke for a minimum duration of 1.5 seconds to get a sugar pellet as a reward. All rats learned to wait longer in the nosepoke when comparing behavior at the beginning and at the end of the experiment. We tested several continuous models against a single change point (CP) model, in which behavior changes at a specific moment and remains stable thereafter. Instead of the traditional approach based on trial-segmented behavior, we used the real time elapsed since the beginning of the experiment as a continuous, uncontrolled variable. We fitted all models to data from individual rats and compared model fit quality across alternatives. Our results provide strong evidence in favor of an abrupt change, as captured by the CP model, over all other models. Moreover, the residuals of the CP model exhibited a Gaussian distribution, suggesting that no additional systematic dynamics remained unexplained and that the behavioral dynamics were fully captured by a single change point.