Fig. 6: Evolution of the weights of states on average, through slow and sudden changes.

Error bars indicate ±1 s.e.m. (lines are slightly offset along the x axis for visibility). Subplots titled by a type represent the weight changes from the first appearance of a state of this type to its last, so only showing state-internal slow changes (and only including states that were present between 5 and 15 sessions, as extremes would skew these averages). Subplots with a title indicating a transition from one type to another show how much each weight of the new state differed from the weights of the closest previously existing state and are based exclusively on the states that first brought the mouse into a new stage. That is, for ‘type 1→2’, we only took into account the first type 2 state exhibited by the mouse and only when that state was type 2 from its inception. For instance, for mouse KS014, this was state 4, which started as type 2 before using the slow process to become type 3. Colored diamond markers on the leftmost and rightmost plots indicate the average value of the weights of the very first state of each mouse and of the dominant state on the last session, respectively. To prevent biases from canceling out across the population, we split the bias weights into the following two groups: starting out below 0 (bias left) or starting out above 0 (bias right). While contrast sensitivities increased both through fast and slow changes, it is noticeable that biases stayed almost constant throughout the lifetime of a state on average, but changed more noticeably through sudden transitions.