The Costs and Benefits of Goal-Directed Attention in Deep Convolutional Neural Networks

People deploy top-down, goal-directed attention to accomplish tasks, such as finding lost keys. By tuning the visual system to relevant information sources, object recognition can become more efficient (a benefit) and more biased toward the target (a potential cost). Motivated by selective attention in categorisation models, we developed a goal-directed attention mechanism that can process naturalistic (photographic) stimuli. Our attention mechanism can be incorporated into any existing deep convolutional neural networks (DCNNs). The processing stages in DCNNs have been related to ventral visual stream. In that light, our attentional mechanism incorporates top-down influences from prefrontal cortex (PFC) to support goal-directed behaviour. Akin to how attention weights in categorisation models warp representational spaces, we introduce a layer of attention weights to the mid-level of a DCNN that amplify or attenuate activity to further a goal. We evaluated the attentional mechanism using photographic stimuli, varying the attentional target. We found that increasing goal-directed attention has benefits (increasing hit rates) and costs (increasing false alarm rates). At a moderate level, attention improves sensitivity (i.e. increases $d^{\prime }$ d ′ ) at only a moderate increase in bias for tasks involving standard images, blended images and natural adversarial images chosen to fool DCNNs. These results suggest that goal-directed attention can reconfigure general-purpose DCNNs to better suit the current task goal, much like PFC modulates activity along the ventral stream. In addition to being more parsimonious and brain consistent, the mid-level attention approach performed better than a standard machine learning approach for transfer learning, namely retraining the final network layer to accommodate the new task.

The Disorder Perceptions of Nonresidents: A Textual Analysis of Open–Ended Survey Responses to Photographic Stimuli

Nonresidents’ perceptions of disorder are potentially consequential for neighborhoods in many ways, as disorder shapes individuals’ behavior within neighborhoods. Unfortunately, there is little research which delves into understanding how nonresidents perceive disorder. Our study provides insight into the perceptions of nonresidents by assessing their interpretations of disorder through their reaction to three photographic stimuli of neighborhoods where they do not live. Through qualitative analysis, we examine various themes in the responses, including disorder theory and both implicit and explicit racial bias. Results show that while nonresidents do have traditional interpretations of disorder, they also interpret disorder in many different ways. Also, even in the absence of people in the photographic stimuli, nonresidents frequently associated disorder with race. Given that nonresidents have the capability to influence the flow of money and resources into the neighborhood, their racially encoded disorder perceptions may have the unintended consequence of entrenching neighborhood issues like segregation, concentrated disadvantage, or unemployment that are common in minority neighborhoods.

Absolute and relative knowledge of ordinal position

For more than 100 years, psychologists have struggled to determine what is learned during serial learning. The method of derived lists is a powerful tool for studying this question. In two experiments, we trained human participants to learn implicit lists by the Transitive Inference (TI) method. We then tested their knowledge of ordinal position of those items. In Experiment 1, participants were presented with pairs of photographic stimuli from five different 5-item training lists by presenting adjacent pairs of items from one list on every trial. Participants were then tested on pairs of items drawn from different lists, in which each item maintained its original ordinal position as it had during training. In Experiment 2, a different group of participants was trained on the same five 5-item lists as that of Experiment 1. However, the order of the items in the derived lists of Experiment 2 was changed systematically. In this latter experiment, the acquisition rate for the derived lists varied inversely with the degree of change of ordinal position. We explain these results by using a model in which participants learn to make positional, as well as transitive inferences, allowing them to infer the relative and absolute position of each item during testing on derived lists.

Absolute and relative knowledge of ordinal position

For more than 100 years, psychologists have struggled to determine what is learned during serial learning. The method of derived lists is a powerful tool for studying this question. In two experiments, we trained human participants to learn implicit lists by the Transitive Inference (TI) method. We then tested their knowledge of ordinal position of those items. In Experiment 1, participants were presented with pairs of photographic stimuli from five different 5-item training lists by presenting adjacent pairs of items from one list on every trial. Participants were then tested on pairs of items drawn from different lists, in which each item maintained its original ordinal position as it had during training. In Experiment 2, a different group of participants was trained on the same five 5-item lists as that of Experiment 1. However, the order of the items in the derived lists of Experiment 2 was changed systematically. In this latter experiment, the acquisition rate for the derived lists varied inversely with the degree of change of ordinal position. We explain these results by using a model in which participants learn to make positional, as well as transitive inferences, allowing them to infer the relative and absolute position of each item during testing on derived lists.

Do Pictures of Faces, and Which Ones, Capture Attention in the Inattentional-Blindness Paradigm?

Faces and self-referential material (eg one's own name) are more likely to capture attention in the inattentional-blindness (IB) paradigm than other stimuli. This effect is presumably due to the meaning of these stimuli rather than to their familiarity [Mack and Rock, 1998 Inattentional Blindness (Cambridge, MA: MIT Press)]. In previous work, IB has been investigated mostly with schematic stimuli. In the present study, the generalisability of this finding was tested with photographic stimuli. In support of the view that faces constitute a special category of stimuli, pictures of faces were found to resist more to IB than pictures of common objects (experiment 1) or than pictures of inverted faces (experiment 2). In a third experiment, the influence of face familiarity and identity (the participant's own face, a friend's face, and an unknown face) on IB rates was evaluated. Unexpectedly, no differential resistence to blindness across these three kinds of faces was found. In conclusion, pictures of faces attracted attention more than pictures of objects or inverted faces in the IB paradigm. However, this effect was not dependent on face familiarity or identity.