Bottom-up Attention Research Articles

Exploring sex differences in auditory saliency: the role of acoustic characteristics in bottom-up attention

BackgroundSeveral cognitive functions are related to sex. However, the relationship between auditory attention and sex remains unclear. The present study aimed to explore sex differences in auditory saliency judgments, with a particular focus on bottom-up type auditory attention.MethodsForty-five typical adults (mean age: 21.5 ± 0.64 years) with no known hearing deficits, intelligence abnormalities, or attention deficits were enrolled in this study. They were tasked with annotating attention capturing sounds from five audio clips played in a soundproof room. Each stimulus contained ten salient sounds randomly placed within a 1-min natural soundscape. We conducted a generalized linear mixed model (GLMM) analysis using the number of responses to salient sounds as the dependent variable, sex as the between-subjects factor, duration, maximum loudness, and maximum spectrum of each sound as the within-subjects factor, and each sound event and participant as the variable effect.ResultsNo significant differences were found between male and female groups in age, hearing threshold, intellectual function, and attention function (all p > 0.05). Analysis confirmed 77 distinct sound events, with individual response rates of 4.0–100%. In a GLMM analysis, the main effect of sex was not statistically significant (p = 0.458). Duration and spectrum had a significant effect on response rate (p = 0.006 and p < 0.001). The effect of loudness was not statistically significant (p = 0.13).ConclusionsThe results suggest that male and female listeners do not differ significantly in their auditory saliency judgments based on the acoustic characteristics studied. This finding challenges the notion of inherent sex differences in bottom-up auditory attention and highlights the need for further research to explore other potential factors or conditions under which such differences might emerge.

Negative auditory hallucinations are associated with increased activation of the defensive motivational system in schizophrenia

Auditory hallucinations (AH) are the most common symptom of psychosis. The voices people hear make comments that are benign or even encouraging, but most often voices are threatening and derogatory. Negative AH are often highly distressing and contribute to suicide risk and violent behavior. Biological mechanisms underlying the valence of voices (i.e., positive, negative, neutral) are not well delineated. In the current study, we examined whether AH voice valence was associated with increased activation of the Defensive Motivational System, as indexed by central and autonomic system response to unpleasant stimuli. Data were evaluated from two studies that used a common symptom rating instrument, the Psychotic Symptom Rating Scale (PSY-RATS), to measure AH valence. Participants included outpatients diagnosed with SZ. Tasks included: Study 1: Trier Social Stress Task while heart rate was recorded via electrocardiography (N = 27); Study 2: Passive Viewing Task while participants were exposed to pleasant, unpleasant, and neutral images from the International Affective Picture System (IAPS) library while eye movements, pupil dilation, and electroencephalography were recorded (N = 25). Results indicated that negative voice content was significantly associated with: 1) increased heart rate during an acute social stressor, 2) increased pupil dilation to unpleasant images, 3) higher neural reactivity to unpleasant images, and 4) a greater likelihood of having bottom-up attention drawn to unpleasant stimuli. Findings suggest that negative AH are associated with greater Defensive Motivational System activation in terms of central and autonomic nervous system response.

Robust Tracking via Combing Top-Down and Bottom-Up Attention

Robust Tracking via Combing Top-Down and Bottom-Up Attention

Developing the Attentive Brain: Contribution of Cognitive Neuroscience to a Theory of Attentional Development

Attention is a fundamental cognitive mechanism whose primary function is to regulate and organize the flow of perceptions and actions shaping our mental life. Early cognitive models have highlighted aspects of sustained, selective, and executive control as essential components of attention. These three broad aspects can be further subdivided into subordinate operations, depending on whether the particular function is mostly driven by external stimulation (bottom-up attention) or relies on endogenous processes such as voluntary intentions or expectations (top-down attention). After several decades of cognitive neuroscience research, these different functions have been associated with specific circuits of brain regions. Based on the cognitive neuroscience framework, this paper presents a theory of attention development and discusses behavioral and brain evidence regarding the development of attention function during the first years of life.

The Frontal Lobe and Top-down Control of Attention

The concept of attention in cognitive science encompasses a bidirectional nature: bottom-up attention based on the salience of sensory stimuli, and top-down attention, which involves voluntary control over aspects such as intensity, allocation, selectivity, and duration. Top-down attention is believed to be primarily realized through the frontal lobes that monitor on-going information processing. This monitoring helps detect situations requiring intervention and manipulates lower-level information processing systems as a part of executive functions.

Studying the Role of Visuospatial Attention in the Multi-Attribute Task Battery II

Task batteries mimicking user tasks are of high heuristic value. Supposedly, they measure individual human aptitude regarding the task in question. However, less is often known about the underlying mechanisms or functions that account for task performance in such complex batteries. This is also true of the Multi-Attribute Task Battery (MATB-II). The MATB-II is a computer display task. It aims to measure human control operations on a flight console. Using the MATB-II and a visual-search task measure of spatial attention, we tested if capture of spatial attention in a bottom-up or top-down way predicted performance in the MATB-II. This is important to understand for questions such as how to implement warning signals on visual displays in human–computer interaction and for what to practice during training of operating with such displays. To measure visuospatial attention, we used both classical task-performance measures (i.e., reaction times and accuracy) as well as novel unobtrusive real-time pupillometry. The latter was done as pupil size covaries with task demands. A large number of analyses showed that: (1) Top-down attention measured before and after the MATB-II was positively correlated. (2) Test-retest reliability was also given for bottom-up attention, but to a smaller degree. As expected, the two spatial attention measures were also negatively correlated with one another. However, (3) neither of the visuospatial attention measures was significantly correlated with overall MATB-II performance, nor with (4) any of the MATB-II subtask performance measures. The latter was true even if the subtask required visuospatial attention (as in the system monitoring task of the MATB-II). (5) Neither did pupillometry predict MATB-II performance, nor performance in any of the MATB-II’s subtasks. Yet, (6) pupil size discriminated between different stages of subtask performance in system monitoring. This finding indicated that temporal segregation of pupil size measures is necessary for their correct interpretation, and that caution is advised regarding average pupil-size measures of task demands across tasks and time points within tasks. Finally, we observed surprising effects of workload (or cognitive load) manipulation on MATB-II performance itself, namely, better performance under high- rather than low-workload conditions. The latter findings imply that the MATB-II itself poses a number of questions about its underlying rationale, besides allowing occasional usage in more applied research.

An Anatomy- and Topology-Preserving Framework for Coronary Artery Segmentation.

Coronary artery segmentation is critical for coronary artery disease diagnosis but challenging due to its tortuous course with numerous small branches and inter-subject variations. Most existing studies ignore important anatomical information and vascular topologies, leading to less desirable segmentation performance that usually cannot satisfy clinical demands. To deal with these challenges, in this paper we propose an anatomy- and topology-preserving two-stage framework for coronary artery segmentation. The proposed framework consists of an anatomical dependency encoding (ADE) module and a hierarchical topology learning (HTL) module for coarse-to-fine segmentation, respectively. Specifically, the ADE module segments four heart chambers and aorta, and thus five distance field maps are obtained to encode distance between chamber surfaces and coarsely segmented coronary artery. Meanwhile, ADE also performs coronary artery detection to crop region-of-interest and eliminate foreground-background imbalance. The follow-up HTL module performs fine segmentation by exploiting three hierarchical vascular topologies, i.e., key points, centerlines, and neighbor connectivity using a multi-task learning scheme. In addition, we adopt a bottom-up attention interaction (BAI) module to integrate the feature representations extracted across hierarchical topologies. Extensive experiments on public and in-house datasets show that the proposed framework achieves state-of-the-art performance for coronary artery segmentation.

Digital Distraction, Attention Regulation, and Inequality

In the popular and academic literature on the problems of the so-called attention economy, the cost of attention grabbing, sustaining, and immersing digital medias has been addressed as if it touched all people equally. In this paper I ask whether everyone has the same resources to respond to the recent changes in their stimulus environments caused by the attention economy. I argue that there are not only differences but disparities between people in their responses to the recent, significant increase in the degree and persuasiveness of digital distraction. I point toward individual variance in an agent’s top-down and bottom-up attention regulation, and to further inequality-exacerbating variance in active participation on the internet and in regulating reward-seeking behaviors on the internet. Individual differences in these areas amount to disparities because they have been found to be connected to socioeconomic background factors. I argue that disparities in responding to digital distraction threaten fair equality of opportunity when it comes to digital distraction in the classroom and that they may lead to an unequal contribution of achievements that require complex cognition by people from lower socioeconomic backgrounds.

USER: Unified Semantic Enhancement With Momentum Contrast for Image-Text Retrieval.

As a fundamental and challenging task in bridging language and vision domains, Image-Text Retrieval (ITR) aims at searching for the target instances that are semantically relevant to the given query from the other modality, and its key challenge is to measure the semantic similarity across different modalities. Although significant progress has been achieved, existing approaches typically suffer from two major limitations: (1) It hurts the accuracy of the representation by directly exploiting the bottom-up attention based region-level features where each region is equally treated. (2) It limits the scale of negative sample pairs by employing the mini-batch based end-to-end training mechanism. To address these limitations, we propose a Unified Semantic Enhancement Momentum Contrastive Learning (USER) method for ITR. Specifically, we delicately design two simple but effective Global representation based Semantic Enhancement (GSE) modules. One learns the global representation via the self-attention algorithm, noted as Self-Guided Enhancement (SGE) module. The other module benefits from the pre-trained CLIP module, which provides a novel scheme to exploit and transfer the knowledge from an off-the-shelf model, noted as CLIP-Guided Enhancement (CGE) module. Moreover, we incorporate the training mechanism of MoCo into ITR, in which two dynamic queues are employed to enrich and enlarge the scale of negative sample pairs. Meanwhile, a Unified Training Objective (UTO) is developed to learn from mini-batch based and dynamic queue based samples. Extensive experiments on the benchmark MSCOCO and Flickr30K datasets demonstrate the superiority of both retrieval accuracy and inference efficiency. For instance, compared with the existing best method NAAF, the metric R@1 of our USER on the MSCOCO 5K Testing set is improved by 5% and 2.4% on caption retrieval and image retrieval without any external knowledge or pre-trained model while enjoying over 60 times faster inference speed. Our source code will be released at https://github.com/zhangy0822/USER.

A lateralized alpha-band marker of the interference of exogenous attention over endogenous attention.

Scientists differentiate top-down from bottom-up visuospatial attention processes. While bottom-up attention is rapidly engaged by emerging demands from the environment, top-down attention in contrast reflects slow voluntary shifts of spatial attention. Several lines of research substantiate the idea that top-down and bottom-up attentional processes involve distinct functional systems. An increasing number of studies, however, argue that both attention systems share brain processing resources. The current study examines how salient visual events that engage bottom-up processes interfere with top-down attentional processes. Using neurophysiological recordings and multivariate pattern classification techniques, the authors show that these patterns of interference occur within the alpha-band of neural activity (8-12Hz), which implies that bottom-up and top-down attention processes share this narrow-band frequency brain resource. The results further demonstrate that patterns of alpha-band activity explains, in part, the interference between top-down and bottom-up attention at the behavioral level.

The Electrophysiological Markers of Statistically Learned Attentional Enhancement: Evidence for a Saliency-based Mechanism.

It is well established that attention can be sharpened through the process of statistical learning (e.g., visual search becomes faster when targets appear at high-relative-to-low probability locations). Although this process of statistically learned attentional enhancement differs behaviorally from the well-studied top-down and bottom-up forms of attention, relatively little work has been done to characterize the electrophysiological correlates of statistically learned attentional enhancement. It thus remains unclear whether statistically learned enhancement recruits any of the same cognitive mechanisms as top-down or bottom-up attention. In the current study, EEG data were collected from while participants searched for an ambiguous unique shape in a visual array (the additional singleton task). Unbeknown to the participants, targets appeared more frequently in one location in space (probability cuing). Encephalographic data were then analyzed in two phases: an anticipatory phase and a reactive phase. In the anticipatory phase preceding search stimuli onset, alpha lateralization as well as the Anterior Directing Attention Negativity and Late Directing Attention Positivity components-signs of preparatory attention known to characterize top-down enhancement-were tested. In the reactive phase, the N2pc component-a well-studied marker of target processing-was examined following stimuli onset. Our results showed that statistically learned attentional enhancement is not characterized by any of the well-known anticipatory markers of top-down attention; yet targets at high probability locations did reliably evoke larger N2pc amplitudes, a finding that is associated with bottom-up attention and saliency. Overall, our findings are consistent with the notion that statistically learned attentional enhancement increases the perceptual salience of items appearing at high-probability locations relative to low-probability locations.

Task-Based Visual Attention for Continually Improving the Performance of Autonomous Game Agents

Deep Reinforcement Learning (DRL) has been effectively performed in various complex environments, such as playing video games. In many game environments, DeepMind’s baseline Deep Q-Network (DQN) game agents performed at a level comparable to that of humans. However, these DRL models require many experience samples to learn and lack the adaptability to changes in the environment and handling complexity. In this study, we propose Attention-Augmented Deep Q-Network (AADQN) by incorporating a combined top-down and bottom-up attention mechanism into the DQN game agent to highlight task-relevant features of input. Our AADQN model uses a particle-filter -based top-down attention that dynamically teaches an agent how to play a game by focusing on the most task-related information. In the evaluation of our agent’s performance across eight games in the Atari 2600 domain, which vary in complexity, we demonstrate that our model surpasses the baseline DQN agent. Notably, our model can achieve greater flexibility and higher scores at a reduced number of time steps.Across eight game environments, AADQN achieved an average relative improvement of 134.93%. Pong and Breakout games both experienced improvements of 9.32% and 56.06%, respectively. Meanwhile, SpaceInvaders and Seaquest, which are more intricate games, demonstrated even higher percentage improvements, with 130.84% and 149.95%, respectively. This study reveals that AADQN is productive for complex environments and produces slightly better results in elementary contexts.

Neuroimaging evidence supporting a dual-network architecture for the control of visuospatial attention in the human brain: a mini review.

Neuroimaging studies conducted in the last three decades have distinguished two frontoparietal networks responsible for the control of visuospatial attention. The present review summarizes recent findings on the neurophysiological mechanisms implemented in both networks and describes the evolution from a model centered on the distinction between top-down and bottom-up attention to a model that emphasizes the dynamic interplay between the two networks based on attentional demands. The role of the dorsal attention network (DAN) in attentional orienting, by boosting behavioral performance, has been investigated with multiple experimental approaches. This research effort allowed us to trace a distinction between DAN regions involved in shifting vs. maintenance of attention, gather evidence for the modulatory influence exerted by the DAN over sensory cortices, and identify the electrophysiological correlates of the orienting function. Simultaneously, other studies have contributed to reframing our understanding of the functions of the ventral attention network (VAN) and its relevance for behavior. The VAN is not simply involved in bottom-up attentional capture but interacts with the DAN during reorienting to behaviorally relevant targets, exhibiting a general resetting function. Further studies have confirmed the selective rightward asymmetry of the VAN, proposed a functional dissociation along the anteroposterior axis, and suggested hypotheses about its emergence during the evolution of the primate brain. Finally, novel models of network interactions explain the expression of complex attentional functions and the emergence and restorations of symptoms characterizing unilateral spatial neglect. These latter studies emphasize the importance of considering patterns of network interactions for understanding the consequences of brain lesions.

Driver's visual fixation attention prediction in dynamic scenes using hybrid neural networks

The traffic driving scenario is dynamically varying and experienced drivers can allocate visual attention effectively in brief time intervals, focusing on prominent targets and areas in advance, thus ensuring safe driving. Therefore, it is significant to research the behavior of driver's visual attention allocation in the development of driver assistance systems and autonomous vehicles. Recently, inspired by top-down and bottom-up attention mechanisms, most visual attention models based on neural networks have been proposed. However, these models tend to yield highly distributed predictions for the specific driving scenario, and have relatively low prediction accuracy, making them unfeasible for being used in practice. Hence, to overcome these challenges, we propose a driver's visual attention model (DVAM) built on the encoder-decoder architecture with the utilization of the convolutional neural network (CNN) and the recurrent neural networks (RNN). Specifically, the model leverages the temporal and spatial dimension information to better mimic realistic dynamic driving, and extract abundant feature information, thus ensuring that the final predictions are authentic and effective. Then, extensive experiments demonstrate that our suggested DVAM achieves better robustness and precision in predicting driver attention to areas or targets compared to state-of-the-art methods on the DR(eye)VE dataset. Finally, our model is deployed on the TDV dataset to validate its generalization capabilities, which prove the potential for field adaptation.

Show me that you are advertising: Visual salience of products attenuates detrimental effects of persuasion knowledge activation in influencer advertising

In the realm of social media influencer advertising, the organic nature of sponsored content can deceive users by concealing persuasive intent. Regulators seek to reinforce users' persuasion knowledge through sponsorship disclosure guidelines, while advertisers explore ethical tactics to prevent the adverse effects of users' ad recognition on the effectiveness of influencer endorsements. With the increasing prevalence of image-based content, the investigation of visual factors, such as the presentation of products, and their moderating influence on the effects of ad recognition becomes ever more relevant. However, previous research has primarily focused on textual factors. Building on congruence theory, we demonstrate that marketers can harness bottom-up attention induced by high visual salience to counteract the detrimental effects of users’ ad recognition. Our findings indicate that ad recognition undermines favorable brand attitudes, word-of-mouth intentions, and purchase intentions by diminishing perceived influencer integrity and enhancing post skepticism. These effects are alleviated when the visual appearance of sponsored products is highly salient in terms of brightness or color. Our work offers a simple yet effective image-based solution for marketers to mitigate ad-related skepticism and elevate the persuasive potential of influencer advertising in the visually-driven realm of computer-mediated communication.

The Effects of In-Stream Video Advertising on Ad Information Encoding: A Neurophysiological Study

Although in-stream video advertising is common, its effects on advertisement (ad) information encoding remain unclear. We investigated the effects of in-stream video advertising by comparing two groups: those watching mid-roll (between the program) ads and those watching pre- and post-roll (before and after the program, respectively) ads. To elucidate how advertising content is encoded in the context of in-stream video advertising, we integrated two theoretical frameworks: the negative emotion–memory model (NEMM) and the limited capacity model of motivated–mediated message processing (LC4MP). We used electroencephalography (EEG) to assess negative emotions and bottom-up attention during advertisement viewing. The findings indicate that the first mid-roll ad induced negative emotions, but these feelings were attenuated during subsequent mid-rolls. In addition, negative emotions induced by mid-roll ads attenuated the role of bottom-up attention in the information encoding process. However, the pre- and post-roll ads were not accompanied by negative emotions; thus, bottom-up attention played a major role in the information encoding of these ads. The results also suggest that despite the negative emotions experienced during mid-rolls, such transient negative reactions did not affect purchase intention for the advertised products.

Control of bottom-up attention in scene cognition contributes to visual working memory performance.

Several studies have investigated the relationship between working memory and attention. However, most of the relevant studies so far investigated top-down attention; only a few have examined possible interactions between bottom-up attention and visual working memory. In the present study, we focused on the visual saliency of different parts of pictures as an index of the degree to which one's bottom-up attention can be drawn towards each of them. We administered the Picture Span Test (PST) to investigate whether salient parts of pictures can influence the performance of visual working memory. The task required participants to judge the semantic congruency of objects in pictures and remember specific parts of pictures. In Experiment 1, we calculated a saliency map for the PST stimuli and found that salient but task-irrelevant parts of pictures could evoke intrusion errors. In Experiment 2, we demonstrated that longer gazing time at target areas results in a higher probability of making correct recognition. In addition, frequent gaze fixation and high normalized scan-path saliency values in task-irrelevant areas were associated with intrusion errors. These results suggest that visual information processed by bottom-up attention may affect working memory.

Interhemispheric competition during sleep

Our understanding of the functions and mechanisms of sleep remains incomplete, reflecting their increasingly evident complexity1–3. Likewise, studies of interhemispheric coordination during sleep4–6 are often hard to connect precisely to known sleep circuits and mechanisms. Here, by recording from the claustra of sleeping bearded dragons (Pogona vitticeps), we show that, although the onsets and offsets of Pogona rapid-eye-movement (REMP) and slow-wave sleep are coordinated bilaterally, these two sleep states differ markedly in their inter-claustral coordination. During slow-wave sleep, the claustra produce sharp-wave ripples independently of one another, showing no coordination. By contrast, during REMP sleep, the potentials produced by the two claustra are precisely coordinated in amplitude and time. These signals, however, are not synchronous: one side leads the other by about 20 ms, with the leading side switching typically once per REMP episode or in between successive episodes. The leading claustrum expresses the stronger activity, suggesting bilateral competition. This competition does not occur directly between the two claustra or telencephalic hemispheres. Rather, it occurs in the midbrain and depends on the integrity of a GABAergic (γ-aminobutyric-acid-producing) nucleus of the isthmic complex, which exists in all vertebrates and is known in birds to underlie bottom-up attention and gaze control. These results reveal that a winner-take-all-type competition exists between the two sides of the brain of Pogona, which originates in the midbrain and has precise consequences for claustrum activity and coordination during REMP sleep.

Effects of top-down and bottom-up attention on post-selection posterior contralateral negativity.

We examined the effect of combined top-down and bottom-up attentional control sources in easy and difficult visual search tasks. Applying a new analysis on previously acquired data, we focused on the sustained posterior contralateral negativity (SPCN) and the response-locked posterior contralateral negativity (RLpcN), to better understand processes following target selection. We used the signed-area approach to measure the negative area, where the signal was either locked to the target or the response onsets. We further split the RLpcN into an early and a late segment to capture the dynamics of selection and post-selection processes. In Experiment 1, participants reported the orientation of a uniquely tilted target. In Experiment 2, participants reported the position of a small gap within the uniquely tilted target. In both experiments, endogenous cues manipulated top-down attention (valid vs. neutral), and salient color singletons (either the target or a distractor) manipulated bottom-up attention. We hypothesized that the SPCN and the later segment of the RLpcN would be modulated by task difficulty and target salience, as they are associated with post-selection processes, such as response selection and working memory. The early segment of the RLpcN was hypothesized to be modulated by the cueing manipulation and presence of a salient distractor, as they affect target selection. An effect of distractor presence was observed on the early segment of the RLpcN, and our results further supported the hypotheses regarding the SPCN and the later segment of the RLpcN, providing novel insights into post-selection processes in visual search.

The effect of visual attention process and thinking styles on environmental aesthetic preference: An eye-tracking study.

People often form different aesthetic preferences for natural and built environments, which affects their behavioral intention; however, it remains unknown whether this difference in aesthetic preference is due to differences in thinking styles. However, whether tourists' aesthetic preferences differ when using different visual attention processes has not been studied further. This study used eye-tracking and self-reporting to investigate these questions. The results show that natural environment images are more favored visually because they can evoke in tourists larger pupil diameters and longer scan paths, but we found no significant difference in fixation duration and fixation counts. We also found that the scanning path of tourists who predominantly rely on intuitive thinking is modulated by the bottom-up attention process, while the scanning path of tourists who prefer rational thinking is modulated by the top-down attention process. In the bottom-up process, tourists who prefer rational thinking exhibit more positive aesthetic preferences and emotional arousal. In summary, the present study verified that aesthetic preference is more likely to be influenced by both thinking style and visual attention processing. The results of the present work provide preliminary evidence that the aesthetic preference of the environment is not only related to visual attention but also affected by the individual visual attention process and thinking style.