Evidence for criterion shifts in visual perceptual learning: Data and implications

Abstract

) documents the range of tasks in which practice on elementary visual judgments can produce dramatic improvements in performance. This behavioral evidence has been used to motivate the development of a range of models for the mechanisms of perceptual learning (e.g., Adini & Sagi, 2001;Adini, Sagi, & Tsodyks, 2002;Gold, Bennett, & Sekuler, 1999;Gold, Sekuler, & Bennett, 2004;Petrov, Dosher, & Lu, 2005;Yu, Klein, & Levi, 2004), with the majority of these models focused on how perceptual sensitivity to either the presence of or a variation in a defining characteristic changes with experience. The work presented here explores the possibility that the changes observed in visual perceptual learning may also reflect changes in response (decisional) criteria for both detection and identification, in addition to or instead of changes in perceptual sensitivity.In order to ask these questions, it is critical to consider the extent to which the behavioral evidence available in any study of perceptual learning can speak in a reasonably restrictive manner to changes in perceptual sensitivity (Fahle & Poggio, 2002;Sagi & Tanne, 1994;Zenger & Sagi, 2002). If that evidence can reflect changes due to factors other than changes in sensitivity-such as changes in biases for detection, identification, response selection, or all three-inferences regarding the underlying mechanisms may be in error. Consider, in particular, the possibility that experience may produce changes in both sensitivity and bias. The latter could suggest important roles for selective attention, memory, or other influences, such as might be available by way of back-projections within the visual system (e.g., Di Lollo, Enns, & Rensink, 2000;Lamme & Roelfsema, 2000;Li, Piëch, & Gilbert, 2004;Ling & Carrasco, 2006;Sigman & Gilbert, 2000;Sigman et al., 2005). However, if the experimental approach does not provide the data necessary to measure both types of potential changes, or if it does not adequately control for the possible effects of different types of response bias, any inferences regarding any single basis for change is potentially suspect.In a large number of the studies on perceptual learning, an experimental approach-an adaptive version of a Work on visual perceptual learning for contrast detection has shown that reliable decreases in detection thresholds are accompanied by reliable increases in false alarm rates (Wenger & Rasche, 2006). The present study assesses the robustness and replicability of these changes, demonstrating that they are independent of a variety of task demands (i.e., the specific method used for perceptual practice and threshold estimation) and the presence or absence of trial-by-trial feedback and that the source of the increases can be found in shifts in changes in sensitivity and in bias for detection, identification, or both. Although the increase in false alarm rates suggests a strategic shift in response criteria for detection, we demonstrate that there are multiple potential explanations, including explanations that do not requi...

Reducing Crowding by Weakening Inhibitory Lateral Interactions in the Periphery with Perceptual Learning

“…Since the yes/no procedure that we have used is sensitive to response bias [37], the procedure may have had a significant impact on the reported thresholds. To check for this possibility, we have reanalyzed the data by calculating d ', which is a measure of sensitivity that is independent of bias.…”

Accounting for speed–accuracy tradeoff in perceptual learning

“…Previous studies have shown that perceptual learning can induce significant changes in response bias, such as an increase in detection false alarms rates (Seitz, Nanez, Holloway, Koyama, & Watanabe, 2005; Wenger & Rasche, 2006; Wenger, Copeland, Bittner, & Thomas, 2008). Thus, we also calculated the response bias according to the SDT criterion, c .…”

Faces in the Mist: Illusory Face and Letter Detection

“…The use of an illusory detection paradigm (i.e., only noise images) introduces a complication in that some observers may produce much higher false alarm rates than other observers (Wenger et al 2008). To the extent that this occurs, it introduces noise in the CIs because some observers respond due to a bias to respond in general rather than respond based on a good match between the noise image and their internal template.…”