Subject best-fit learning model trial-by-trial analysis.

(a) Example of trial-by-trial energy landscape changes (top) and log-likelihood ratio between best learning model compared to the best fixed heterogeneous model (bottom) in subject with the same best-fit learning model for both short and long delays. Red dashed line shows where learning qualitatively overcomes the initial biases in the landscape and corresponds to an increase in LLR trials in favor of the learning model. (b) Total LLR summed across trials for all subjects that were consistently best matched to the same learning model (top) and the fraction of subjects with a total LLR that is positive across trials (bottom). We see that most subjects increase their total LLR over time, suggesting that learning models are becoming more aligned with subjects’ responses. (TIF)