Optimal classifier feedback improves cost-benefit but not base-rate decision criterion learning in perceptual categorization

Targeting separate specific learning parameters underlying cognitive behavioral therapy can improve perceptual judgments of anger

BACKGROUND AND OBJECTIVES

Anxiety disorders are characterized by biased perceptual judgment. An experimental model using simple verbal instruction to target specific decision parameters that influence perceptual judgment was developed to test if it could influence anger perception, and to examine differences between individuals with social anxiety disorder (SAD) relative to generalized anxiety disorder (GAD) or non-psychiatric controls.

METHODS

Anger perception was decomposed into three decision parameters (perceptual similarity of angry vs. not-angry facial expressions, base rate of encountering angry vs. not-angry expressions, payoff for correct vs. incorrect categorization of face stimuli) using a signal detection framework. Participants with SAD (n = 97), GAD (n = 90), and controls (n = 98) were assigned an instruction condition emphasizing one of the three decision parameters. Anger perception pre-vs. post-instruction and its interaction with diagnosis were examined.

RESULTS

For all participants, base rate instructions impacted response bias over and above practice effects, supporting the validity of this instructional task-based approach to altering response bias. We failed to find a similarity or payoff instruction effect, nor a diagnosis interaction.

LIMITATIONS

Future instructional tasks may need to more closely target core cognitive and perceptual biases in anxiety disorders to identify specific deficits and how to optimally influence them.

CONCLUSIONS

This study demonstrates that specific decision parameters underlying perceptual judgment can be experimentally manipulated. Although our study failed to show diagnosis specific effects, it suggests that individual parameter "estimation" deficits may be experimentally isolated and potentially targeted, with the ultimate goal of developing an objective approach to personalized intervention targeting biased perceptual judgments in anxiety disorders.

Suboptimality in perceptual decision making

Human perceptual decisions are often described as optimal. Critics of this view have argued that claims of optimality are overly flexible and lack explanatory power. Meanwhile, advocates for optimality have countered that such criticisms single out a few selected papers. To elucidate the issue of optimality in perceptual decision making, we review the extensive literature on suboptimal performance in perceptual tasks. We discuss eight different classes of suboptimal perceptual decisions, including improper placement, maintenance, and adjustment of perceptual criteria; inadequate tradeoff between speed and accuracy; inappropriate confidence ratings; misweightings in cue combination; and findings related to various perceptual illusions and biases. In addition, we discuss conceptual shortcomings of a focus on optimality, such as definitional difficulties and the limited value of optimality claims in and of themselves. We therefore advocate that the field drop its emphasis on whether observed behavior is optimal and instead concentrate on building and testing detailed observer models that explain behavior across a wide range of tasks. To facilitate this transition, we compile the proposed hypotheses regarding the origins of suboptimal perceptual decisions reviewed here. We argue that verifying, rejecting, and expanding these explanations for suboptimal behavior - rather than assessing optimality per se - should be among the major goals of the science of perceptual decision making.

Base-rate sensitivity through implicit learning

Two experiments assessed the contributions of implicit and explicit learning to base-rate sensitivity. Using a factorial design that included both implicit and explicit learning disruptions, we tested the hypothesis that implicit learning underlies base-rate sensitivity from experience (and that explicit learning contributes comparatively little). Participants learned to classify two categories of simple stimuli (bar graph heights) presented in a 3:1 base-rate ratio. Participants learned either from “observational” training to disrupt implicit learning or “response” training which supports implicit learning. Category label feedback on each trial was followed either immediately or after a 2.5 second delay by onset of a working memory task intended to disrupt explicit reasoning about category membership feedback. Decision criterion values were significantly larger following response training, suggesting that implicit learning underlies base-rate sensitivity. Disrupting explicit processing had no effect on base-rate learning as long as implicit learning was supported. These results suggest base-rate sensitivity develops from experience primarily through implicit learning, consistent with separate learning systems accounts of categorization.

Working memory capacity is associated with optimal adaptation of response bias to perceptual sensitivity in emotion perception

Emotion perception, inferring the emotional state of another person, is a frequent judgment made under perceptual uncertainty (e.g., a scowling facial expression can indicate anger or concentration) and behavioral risk (e.g., incorrect judgment can be costly to the perceiver). Working memory capacity (WMC), the ability to maintain controlled processing in the face of competing demands, is an important component of many decisions. We investigated the association of WMC and anger perception in a task in which "angry" and "not angry" categories comprised overlapping ranges of scowl intensity, and correct and incorrect responses earned and lost points, respectively. Participants attempted to earn as many points as they could; adopting an optimal response bias would maximize decision utility. Participants with higher WMC more optimally tuned their anger perception response bias to accommodate their perceptual sensitivity (their ability to discriminate the categories) than did participants with lower WMC. Other factors that influence response bias (i.e., the relative base rate of angry vs. not angry faces and the decision costs and benefits) were ruled out as contributors to the WMC-bias relationship. Our results suggest that WMC optimizes emotion perception by contributing to perceivers' ability to adjust their response bias to account for their level of perceptual sensitivity, likely an important component of adapting emotion perception to dynamic social interactions and changing circumstances. (PsycINFO Database Record

Dubious decision evidence and criterion flexibility in recognition memory

When old–new recognition judgments must be based on ambiguous memory evidence, a proper criterion for responding “old” can substantially improve accuracy, but participants are typically suboptimal in their placement of decision criteria. Various accounts of suboptimal criterion placement have been proposed. The most parsimonious, however, is that subjects simply over-rely on memory evidence – however faulty – as a basis for decisions. We tested this account with a novel recognition paradigm in which old–new discrimination was minimal and critical errors were avoided by adopting highly liberal or conservative biases. In Experiment 1, criterion shifts were necessary to adapt to changing target probabilities or, in a “security patrol” scenario, to avoid either letting dangerous people go free (misses) or harming innocent people (false alarms). Experiment 2 added a condition in which financial incentives drove criterion shifts. Critical errors were frequent, similar across sources of motivation, and only moderately reduced by feedback. In Experiment 3, critical errors were only modestly reduced in a version of the security patrol with no study phase. These findings indicate that participants use even transparently non-probative information as an alternative to heavy reliance on a decision rule, a strategy that precludes optimal criterion placement.

Best-classifier feedback in diagnostic classification training

Diagnostic classification training requires viewing many examples along with category membership feedback. "Objective" feedback based on category membership suggests that perfect accuracy is attainable when it may not be (e.g., with confusable categories). Previous work shows that feedback based on an "optimal" responder (that sometimes makes classification errors) leads to higher long-run reward, especially in unequal category payoff conditions. In the current study, participants learned to classify normal or cancerous mammography images, earning more points for correct "cancer" than "normal" responses. Feedback was either objective or based on performance of an empirically determined "best" classifier. This approach is necessary because theoretically optimal responses cannot be determined with complex real-world stimuli with unknown perceptual distributions. Replicating earlier work that used simple artificial stimuli, we found that best-classifier performance led to decision-criterion values (β) closer to the reward-maximizing criterion, along with higher point totals and a slight reduction (as predicted) in overall accuracy.

"Utilizing" signal detection theory

What do inferring what a person is thinking or feeling, judging a defendant's guilt, and navigating a dimly lit room have in common? They involve perceptual uncertainty (e.g., a scowling face might indicate anger or concentration, for which different responses are appropriate) and behavioral risk (e.g., a cost to making the wrong response). Signal detection theory describes these types of decisions. In this tutorial, we show how incorporating the economic concept of utility allows signal detection theory to serve as a model of optimal decision making, going beyond its common use as an analytic method. This utility approach to signal detection theory clarifies otherwise enigmatic influences of perceptual uncertainty on measures of decision-making performance (accuracy and optimality) and on behavior (an inverse relationship between bias magnitude and sensitivity optimizes utility). A "utilized" signal detection theory offers the possibility of expanding the phenomena that can be understood within a decision-making framework.

Ventral striatum and the evaluation of memory retrieval strategies

Adaptive memory retrieval requires mechanisms of cognitive control that facilitate the recovery of goal-relevant information. Frontoparietal systems are known to support control of memory retrieval. However, the mechanisms by which the brain acquires, evaluates, and adapts retrieval strategies remain unknown. Here, we provide evidence that ventral striatal activation tracks the success of a retrieval strategy and correlates with subsequent reliance on that strategy. Human participants were scanned with fMRI while performing a lexical decision task. A rule was provided that indicated the likely semantic category of a target word given the category of a preceding prime. Reliance on the rule improved decision-making, as estimated within a drift diffusion framework. Ventral striatal activation tracked the benefit that relying on the rule had on decision-making. Moreover, activation in ventral striatum correlated with a participant's subsequent reliance on the rule. Taken together, these results support a role for ventral striatum in learning and evaluating declarative retrieval strategies.

Striatal contributions to declarative memory retrieval

Declarative memory is known to depend on the medial temporal lobe memory system. Recently, there has been renewed focus on the relationship between the basal ganglia and declarative memory, including the involvement of striatum. However, the contribution of striatum to declarative memory retrieval remains unknown. Here, we review neuroimaging and neuropsychological evidence for the involvement of the striatum in declarative memory retrieval. From this review, we propose that, along with the prefrontal cortex (PFC), the striatum primarily supports cognitive control of memory retrieval. We conclude by proposing three hypotheses for the specific role of striatum in retrieval: (1) striatum modulates the re-encoding of retrieved items in accord with their expected utility (adaptive encoding), (2) striatum selectively admits information into working memory that is expected to increase the likelihood of successful retrieval (adaptive gating), and (3) striatum enacts adjustments in cognitive control based on the outcome of retrieval (reinforcement learning).

A direct test of the differentiation mechanism: REM, BCDMEM, and the strength-based mirror effect in recognition memory

We explore competing explanations for the reduction in false alarm rate observed when studied items are strengthened. Some models, such as Retrieving Effectively from Memory (REM; Shiffrin & Steyvers, 1997), attribute the false alarm rate reduction to differentiation, a process in which strengthening memory traces at study directly reduces the memory evidence for lure items. Models with no differentiation mechanism, such as the Bind-Cue-Decide Model of Episodic Memory (BCDMEM; Dennis & Humphreys, 2001), explain the false alarm rate reduction in terms of the strength of items expected at retrieval. To contrast these explanations, we separately manipulated item strength at encoding and retrieval. Participants studied mixed lists of weak and strong items. Weak items were always presented once. On separate lists, strong items were either presented twice (Strong-2X) or five times (Strong-5X). Within each strength condition, participants completed separate tests with mixed (strong and weak) targets, pure weak targets, or pure strong targets. They were correctly informed of the type of target on each test. Results showed that false alarm rates decreased from the strong-2X condition to the strong-5X condition for the mixed and pure-strong tests, but not for the pure-weak tests. That is, false alarm rates were determined by the strength of targets appearing on the test, not by the content of the study list. The results support BCDMEM's expectation-based explanation and not REM's differentiation-based explanation.