Abstract
Humans often make decisions based on uncertain sensory information. Signal detection theory describes detection and discrimination decisions as a comparison of stimulus “strength” to a fixed decision criterion. How is the criterion set? We examine how observers learn to set a decision criterion in an orientation-discrimination task. To investigate mechanisms underlying trial-by-trial criterion placement, we compared two tasks. (1) The typical covert-criterion task: Observers make binary discrimination decisions and the underlying criterion is unobservable. (2) A novel overt-criterion task: Observers explicitly set the decision criterion. In each task, stimuli were ellipses with principle orientations drawn from two categories: Gaussian distributions with different means and equal variance. In the covert-criterion task, observers categorized an ellipse. In the overt-criterion task, observers adjusted the orientation of a line on every trial, which served as the discrimination criterion for the subsequently presented ellipse. Feedback was provided at the end of every trial and the observer’s goal was to maximize the number of correct categorizations. The category means were constant throughout each block but changed across blocks. Observers had to relearn the categories for each block. We compared observer performance to the ideal Bayesian model and several suboptimal models (moving average, exponential moving average, reinforcement learning, and limited memory) that varied in both computational and memory demands. While observers were able to learn the optimal criterion over many trials, we found that, in both tasks, observers used suboptimal learning rules. A model in which the recent history of past samples determines a belief about category means (the exponential moving average rule) fit the data best for most observers and on average. We also investigate an analogous task in which category means change slowly over time (a random walk). Our results reveal dynamic adjustment of discrimination criterion, even after prolonged training.
Meeting abstract presented at VSS 2015