Bayesian accounts of covert selective attention: A tutorial review

Vincent, Benjamin T. · 2015 · Crossref

DOI: 10.3758/s13414-014-0830-0

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This tutorial review examines covert selective attention through the lens of Bayesian inference and optimal observer models. The author addresses the theoretical debate regarding whether attention is a causal mechanism that enhances sensory processing or a by-product of adaptive, rational decision-making. Motivated by the limitations of traditional reaction time paradigms, which confound information processing with speed-accuracy trade-offs, the paper advocates for performance-based paradigms using brief display durations. These methods isolate information processing mechanisms by eliminating eye movements and strategic response timing, allowing for precise quantitative comparison between human behavior and theoretical ideals. The paper details the application of probabilistic graphical models to four specific covert search tasks: cued and uncued versions of yes/no detection and spatial alternative forced choice (localization). The methodology involves defining a "forward model" that describes the generative structure of the stimulus environment, including prior probabilities of display types and the likelihood of sensory observations given noise. The observer is modeled as solving the "inverse problem" by combining these priors with noisy sensory data to compute a posterior distribution of belief. The review distinguishes between general Bayesian inference and optimal observer models, the latter of which assume the observer possesses precise knowledge of environmental statistics and observation noise variances to maximize decision accuracy. The author provides specific mathematical formulations and graphical model notations to illustrate how cues update prior probabilities and how decisions are derived from posterior beliefs. The primary finding is that many experimental phenomena attributed to covert attention, such as cueing effects, emerge naturally as by-products of observers conducting Bayesian inference with noisy sensory observations and prior expectations. The review demonstrates that optimal observer models can accurately predict human performance across a range of experimental manipulations without invoking a separate attentional mechanism that alters sensory encoding precision. Instead, cueing effects are explained by decision-level mechanisms, specifically changes in response thresholds or priors based on the statistical structure of the environment. The paper highlights that while Signal Detection Theory offers similar explanations, Bayesian models provide singular predictions based on posterior beliefs and explicitly account for uncertainty in sensory measurements. The significance of this work lies in reframing attention as a set of adaptive behaviors rather than a distinct cognitive mechanism. By showing that human performance aligns with optimal Bayesian inference, the paper supports the view that observers are adaptively rational, optimizing their behavior to suit environmental constraints. This approach provides a unified theoretical framework for understanding covert attention, bridging the gap between model theory and practical implementation through accessible graphical models and supplementary code. It suggests that discrepancies between human and optimal performance offer clues to further hypotheses, rather than invalidating the Bayesian approach entirely.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• attention selective divided
• visual search