Neuronal responses depend on the present stimulus, the previous stimulus, and consistent alternatives to the present stimulus in a task with stimuli and K = 4 transitions possibilities and two rules.

Top: Activity (arbitrary units) of populations P₁ (green) and P₂ (red) as well as the total activity (black) of all pyramidal neurons. After 1500 presentation steps, the transition rule switches from rule 1 to rule 2. Each presentation step corresponds to the exposure to one stimulus for 100ms. Middle: Spike trains of pyramidal neurons during one presentation step, at different points during learning (from left to right): at the beginning (label 1) and end of the first episode with rule 1 (label 2) and beginning (label 3) and end of the first episode with rule 2 (label 4). If the observation is stronger than the prediction neurons in population P₂ fire (blue dots); whereas if the observation is weaker than the prediction neurons in population P₁ fire (red dots). Pyramidal neurons (16 per stimulus, 8 neurons each from P₁ and P₂) have been sorted according to stimulus numbers for visual clarity. Bottom. Matrix of transitions between stimuli decoded from the weights onto pyramidal neurons. At the end of the first presentation step after a change point (label 3), a new element (red arrow) has appeared in the transition matrix corresponding to the newly observed transition, R_n−1 → R_n. After some time with the novel rule, the new transition matrix is learned (label 4) and the old one is suppressed.