How does trial history contribute to trial-to-trial variability?
Scientific context: Trials matched for stimuli exhibit variability in behavior. Stimulus and trial outcome history (behavior and neural activity) may explain a portion of this variance.
IBL dataset (Behavioral data)
Logistic regression; general linear model; reinforcement learning; decision making Logistic regression to predict the response of the current trial using trial history information
The previous several (e.g., 10) trials’ reward unequivalently affect the response in the current trial, closer trials have heavier weights than further ones. The current trial with repeated stimulus features from the last trial would improve mice’s RT and accuracy.
Previous trials have similar effect on the current trial. The current trial with similar stimulus features (e.g., contrast 70-30 vs 80-20) from the last trial would improve RT and accuracy.
Trial history indeed contributes to trial-to-trial variability. The response of the current trial is a combination of current decision-making and accumulations of past-experiences. The trial history of the stimulus physical properties have a strong modulation on mice’s performance.