Modulations of saliency signals at two hierarchical levels of priority computation revealed by spatial statistical distractor learning.

Abstract

Many attention theories assume that selection is guided by a preattentive, spatial representation of the scene that combines bottom-up stimulus information with top-down influences (task goals and prior experience) to code for potentially relevant locations (priority map). At which level(s) of priority computation top-down influences modulate bottom-up stimulus signals is an open question. In a visual-search task, here we induced experience-driven spatial suppression (statistical learning) by presenting 1 of 2 salient distractors more frequently in one display region than the other. When a distractor standing out in the same dimension as the target was spatially biased in Experiment 1, processing of both the target and another, spatially unbiased distractor standing out in a different dimension was likewise hampered in the suppressed region. This indicates that constraining spatial suppression to a specific distractor feature is not possible, and participants instead resort to purely space-based (distractor-feature-independent) suppression at a supradimensional, overall-priority map. In line with a common locus of suppression, a novel computational model of distraction in visual search captures all 3 location effects with a single spatial-weighting parameter. In contrast, when the different-dimension distractor was spatially biased in Experiment 2, processing of other objects in the suppressed region was unaffected, indicating suppression constrained to a subordinate, dimension-specific level of priority computation. In sum, we demonstrate experience-driven top-down modulations of saliency signals at the overall-priority and dimension-specific levels that do not reach down to the specific distractor features.

Statistical learning of frequent distractor locations in visual search involves regional signal suppression in early visual cortex

“…This differential interference effect was significant ( t (31) = -12.2, p < .001). Thus, even though the two types of distractor were balanced in terms of bottom-up saliency, same-dimension distractors caused considerably more RT interference than different-dimension distractors – in line with previous findings (e.g., Sauter et al 2018, 2019; Liesefeld and Müller 2020).…”

“…Consistent with our previous behavioural studies, this cost effect was more marked, at least numerically, with same-dimension distractors. [In previous studies, there was either no cost with different-dimension distractors (e.g., Liesefeld and Müller 2020), or there was some cost initially, which, however, disappeared over extended task practice (Zhang et al 2019). ]…”

Inter-trial effects in priming of pop-out: Comparison of computational updating models

“…From a theoretical perspective, one final issue to be discussed concerns the relations between target feature priming (PoP), which is characterized by long-lasting inter-trial effects, and dimension priming (e.g., [ 5 ]), which is characterized by relatively short-lived inter-trial effects (modeled in our previous paper [ 12 ]). We believe that, due to the ‘nature’ of the displays typically used in the two paradigms (PoP: three or four regularly arranged items that are relatively widely spaced, with the target being defined in a fixed dimension; dimension priming: multi-item arrays of densely arranged items, maximizing local target feature contrast, with the target-defining dimension being variable), dimension priming impacts the computation of target saliency (which is thought to be dimension-specific in nature; see also [ 48 ]), with minimal involvement of top-down feature-based processes (see, e.g., [ 49 , 50 ]). Specifically, our dimension-weighting account (e.g., [ 5 , 6 ]) assumes that dimensional priming is implemented in terms of the competitive up-/down-modulation of the weights of feature contrast signals computed within the alternative, potentially target-defining dimensions in their transfer to the supra-dimensional ‘priority’ map, which then drives the allocation of attention.…”

Massive Effects of Saliency on Information Processing in Visual Working Memory

“…Many theories of visual search (e.g., Duncan & Humphreys, 1989; Fecteau & Munoz, 2006; Liesefeld & Müller, 2019b, 2021; Wolfe, 2021) assume a preattentive spatial representation of the visual scene coding for relevance at each location and informing a second, attentive-processing stage. This assumption is needed to explain how second-stage focal attention can be allocated to the most promising objects in view without analyzing each object in detail first.…”