Discriminating contrast discontinuities: asymmetries, dipper functions, and perceptual learning

Abstract

In a visual search task, a target has to be found among distractors. For two given elements A and B, the search difficulty can depend on which of the two elements is defined as the target, a phenomenon called search asymmetry. Here, we study to what degree an element's ability to 'win' in a search asymmetry depends on its absolute contrast (first-stage signal) and to what degree it depends on its contrast difference from the background (second-stage signal). One quadrant contained a target texture (2 x 2 Gabor patches of contrast c(tg)), and the other three quadrants contained distractor textures (2 x 2 Gabor patches of contrast c(dt)). These four 'foreground textures' were embedded in a background texture consisting of patches with contrast c(bg). The task was to identify which quadrant contained the target. Quadrants are referred to as increments (foreground contrast c(fg)>c(bg)), or decrements (c(fg)<c(bg)). We found that the second-stage signal determines which element wins the performance asymmetry, i.e. it is easier to find strong increments (decrements) among weak increments (decrements) than vice versa. A comparison of our data with the prediction of the independent-processing model [Vision Res. 40 (2000) 2677] shows that the observed performance asymmetries are in general too large to be attributed to noise differences alone. Rather, asymmetries might reflect a global competition between salient elements. Moreover, performance asymmetries can reverse during practice. We characterize a dipper-shaped nonlinearity on the second stage: discrimination of increment (decrement) signals x and x+Delta x first improves for increasing x, and then deteriorates.

Enabling identification of component processes in perceptual learning with nonparametric hierarchical Bayesian modeling

“…Although most perceptual learning studies have primarily focused on general learning through practice or training, some studies have identified additional component processes. These include between-session consolidation ( Karni, Tanne, Rubenstein, Askenasy, & Sagi, 1994 ; McDevitt, Rokem, Silver, & Mednick, 2014 ; Sasaki & Watanabe, 2015 ; Stickgold et al., 2002 ; Tamaki, Berard, et al., 2020 ; Tamaki, Wang, Watanabe, & Sasaki, 2019 ; Tamaki, Wang, et al., 2020 ; Yotsumoto et al., 2009 ), offline gain ( Bang et al., 2018 ; Shibata et al., 2017 ), between-session forgetting ( Beard, Levi, & Reich, 1995 ; Mascetti et al., 2013 ), within-session adaptation ( Censor, Karni, & Sagi, 2006 ; Censor, Harris, & Sagi, 2016 ; Sagi, 2011b ), and deterioration ( Dosher & Lu, 2020 ; Zenger-Landolt & Fahle, 2001 ). For example, Yang et al.…”

Estimating the Trial-by-Trial Learning Curve in Perceptual Learning with Hierarchical Bayesian Modeling

“…It not only provides a depiction of the time course of learning but also assesses specificity and transfer of learning (see chapter 3 of Dosher & Lu, 2020). It has also been used to evaluate the functional form of learning (e.g., power or exponential function) Dosher & Lu, 2007) and, in some cases, contributions of multiple component processes, such as betweensession improvements due to consolidation (Karni et al, 1994;McDevitt et al, 2014;Sasaki & Watanabe, 2015;Stickgold et al, 2002;Tamaki et al, 2019;Tamaki, Wang, et al, 2020;Yotsumoto et al, 2009) or so-called off-line gain (Bang et al, 2018;Shibata et al, 2017), or alternatively between-session losses due to forgetting (Beard et al, 1995;Mascetti et al, 2013); and/or within-session performance reductions due to processes of adaptation (Censor et al, 2006(Censor et al, , 2016Sagi, 2011a) or general deterioration associated with fatigue or inattention (Dosher & Lu, 2020;Zenger-Landolt & Fahle, 2001).…”