Suboptimal Criterion Learning in Static and Dynamic Environments

Abstract

Humans often make decisions based on uncertain sensory information. Signal detection theory (SDT) describes detection and discrimination decisions as a comparison of stimulus “strength” to a fixed decision criterion. However, recent research suggests that current responses depend on the recent history of stimuli and previous responses, suggesting that the decision criterion is updated trial-by-trial. The mechanisms underpinning criterion setting remain unknown. Here, we examine how observers learn to set a decision criterion in an orientation-discrimination task under both static and dynamic conditions. To investigate mechanisms underlying trial-by-trial criterion placement, we introduce a novel task in which participants explicitly set the criterion, and compare it to a more traditional discrimination task, allowing us to model this explicit indication of criterion dynamics. In each task, stimuli were ellipses with principal orientations drawn from two categories: Gaussian distributions with different means and equal variance. In the covert-criterion task, observers categorized a displayed ellipse. In the overt-criterion task, observers adjusted the orientation of a line that served as the discrimination criterion for a subsequently presented ellipse. We compared performance to the ideal Bayesian learner and several suboptimal models that varied in both computational and memory demands. Under static and dynamic conditions, we found that, in both tasks, observers used suboptimal learning rules. In most conditions, a model in which the recent history of past samples determines a belief about category means fit the data best for most observers and on average. Our results reveal dynamic adjustment of discrimination criterion, even after prolonged training, and indicate how decision criteria are updated over time.

Individual difference in serial dependence results from opposite influences of perceptual choices and motor responses

“…This has been established for various 6 attributes, from basic features including orientation (Cicchini, Mikellidou, & Burr, 2017;7 Fischer & Whitney, 2014; Fritsche, Mostert, & de Lange, 2017) and numerosity (Cicchini,8 Anobile, & Burr, 2014) to face identity (Liberman, Fischer, & Whitney, 2014). Using 9 categorization tasks, positive serial dependence was found for attributes such as orientation 10 (Cicchini et al 2017;Norton, Fleming, & Daw, 2017), motion (Alais, Leung, & Van der Burg, 11 2017) and gender of face (Taubert, Alais, & Burr, 2016), while a repulsive serial dependence 12 was found for motion-induced orientation (Alais et al, 2017), facial expression (Taubert et al, 13 2016). These studies investigated serial dependence by examining how the previous stimulus 14 influences current perception, but another way to look at serial dependence is to examine how 15 previous responses affect current perceptual decision making.…”

A Normative Bayesian Model of Classification for Agents with Bounded Memory

“…First, in categorization tasks, history effects occur as the likelihoods (Norton et al, 2017) implicated in sensory acquisition and discrimination rather than motor control.…”

Human online adaptation to changes in prior probability

“…category A and was clockwise of the criterion line or if the ellipse belonged to category 66 B and was counter-clockwise of the criterion line. The overt-criterion task is an 67 explicit version of the covert-criterion task developed by Norton et al [16]. The 68 overt-criterion task provides a richer dataset than the covert-criterion task in that it 69 affords a continuous measure and allows us to see trial by trial changes in the reported 70 decision criterion, at the expense of being a more cognitive task.…”

“…Most 464 notably, for all except the RL model we assumed knowledge of the category 465 distributions. However, Norton et al [16] found that for the same 466 orientation-categorization task, category means were estimated dynamically, even after 467…”

“…Overt-criterion task In the overt-criterion task, observers completed an explicit 562 version of the categorization task described above that was developed by Norton et 563 al. [16]. At the beginning of each trial (N trials = 800), a line was displayed at the 564 center of the screen.…”