Fig. 1: Reversal learning task and conditions.
From: Interacting with volatile environments stabilizes hidden-state inference and its brain signatures
a Generative stimulus distributions used for the two categories A and B. Stimuli correspond to bars of orientations \(\left\{ {\theta _1, \ldots ,\theta _n} \right\}\) drawn from one of two color-labeled categories, associated with overlapping probability density functions centered on orthogonal orientations. b Trial description. Each trial t consists of a sequence of two to eight stimuli \(\left\{ {\theta _1, \ldots ,\theta _n} \right\}\) drawn from one of the two categories, presented at an average rate of 2 Hz, after which participants provide a response \(r_t\) (the left or right key press) regarding the hidden state of the task being tracked. c Graphical description of cue-based and outcome-based conditions. Left: in the cue-based (Cb) condition, participants are instructed to monitor the deck (category A or B) from which presented cards (oriented stimuli) are drawn. This hidden state \(s_t\) of the task alternates occasionally and unpredictably between trials. In this example, the hidden state \(s_t\) reverses at the end of trial t. Right: in the outcome-based (Ob) condition, the same participants are instructed to select the action (the left or right key press), which draws cards from a target deck. The deck drawn at trial \(t\) depends not only on the hidden state \(s_t\) but also on the response \(r_{t - 1}\) provided at the previous trial. As in the cue-based condition, the hidden state \(s_t\) (the target deck-drawing action in this case) alternates occasionally and unpredictably between trials. In this example, the hidden state \(s_t\) reverses at the end of trial t.
