Fig. 3: The Bayesian model reproduces timing-related phenomena.
From: Reconciling time and prediction error theories of associative learning
a Discontinuous learning curves. In accordance with Fig. 1b, we plot the cumulative responses of a Bayesian agent during the delay-conditioning protocol described in Fig. 1a. b Timescale invariance of response profiles. Response rates of the Bayesian agent for different values of cue-reward interval T (top) and after normalizing the response rate by the maximum response rate and re-scaling time by T (bottom). c Timescale invariant learning. The number of trials to acquisition, nacq, for the Bayesian agent are plotted against C/T for different values of the relative log-prior (εc). Note the log-log scale. d Effect of prior rewards. In the Bayesian model, increasing the number of contextual trials results in the agent requiring more cues to form the association. See also Supplementary Fig. 1a.
