Low-level Attention Research Articles

Pairwise Guided Multilayer Cross-Fusion Network for Bird Image Recognition

Bird identification is the first step in collecting data on bird diversity and abundance, which also helps research on bird distribution and population measurements. Most research has built end-to-end training models for bird detection task via CNNs or attentive models, but many perform unsatisfactorily in fine-grained bird recognition. Bird recognition tasks are highly influenced by factors, including the similar appearance of different subcategories, diverse bird postures, and other interference factors such as tree branches and leaves from the background. To tackle this challenge, we propose the Progressive Cross-Union Network (PC-Net) to capture more subtle parts with low-level attention maps. Based on cross-layer information exchange and pairwise learning, the proposed method uses two modules to improve feature representation and localization. First, it utilizes low- and high-level information for cross-layer feature fusion, which enables the network to extract more comprehensive and discriminative features. Second, the network incorporates deep semantic localization to identify and enhance the most relevant regions in the images. In addition, the network is designed with a semantic guidance loss to improve its generalization for variable bird poses. The PC-Net was evaluated on an extensively used birds dataset (CUB-200-2011), which contains 200 birds subcategories. The results demonstrate that the PC-Net achieved an impressive recognition accuracy of 89.2%, thereby outperforming maintained methods in bird subcategory identification. We also achieved competitive results on two other datasets with data on cars and airplanes. The results indicated that the PC-Net improves the accuracy of diverse bird recognition, as well as other fine-grained recognition scenarios.

A Case Study on Autism Spectrum Disorder with Low Level Attention

Autism spectrum disorder (ASD) is now more widely recognised in both the public and medical spheres. Developmental impairment known as autism spectrum disorder (ASD) is brought on by variations in the brain. Individuals with ASD may struggle with confined or repetitive activities or interests, as well as social communication and engagement. Moreover, people with ASD may learn, move, or pay attention in various ways. Due to the increased number of children and teenagers seeking earlier guidance, evaluation, and diagnosis, there is a strong need for services. All clinicians working with children, including primary care teams, allied healthcare professionals, educational and social care staff, must have a solid understanding of the presentation and assessment of autism spectrum disorders (ASD) and an understanding of the co-morbidities to maximise detection and minimise harm. There are standards in the identification, referral, and diagnosis of autism, even if entry points for referrals can change due to a variety of presentation and local service provision. In order to improve a child's potential, provide appropriate support, and implement focused interventions for ASD and co-occurring problems with the hope of improving outcomes, early identification is advantageous. This case study discusses autism spectrum disorder (ASD) diagnosis and offers professionals who come across a child with suspected ASD an assessment strategy.

Deep Temporal Model-Based Identity-Aware Hand Detection for Space Human-Robot Interaction.

Hand detection is a crucial technology for space human-robot interaction (SHRI), and the awareness of hand identities is particularly critical. However, most advanced works have three limitations: 1) the low detection accuracy of small-size objects; 2) insufficient temporal feature modeling between frames in videos; and 3) the inability of real-time detection. In the article, a temporal detector (called TA-RSSD) is proposed based on the SSD and spatiotemporal long short-term memory (ST-LSTM) for real-time detection in SHRI applications. Next, based on the online tubelet analysis, a real-time identity-awareness module is designed for multiple hand object identification. Several notable properties are described as follows: 1) the hybrid structure of the Resnet-101 and the SSD improves the detection accuracy of small objects; 2) three-level feature pyramidal structure retains rich semantic information without losing detailed information; 3) a group of the redesigned temporal attentional LSTM (TA-LSTM) is utilized for three-level feature map modeling, which effectively achieves background suppression and scale suppression; 4) low-level attention maps are used to eliminate in-class similarity between hand objects, which improves the accuracy of identity awareness; and 5) a novel association training scheme enhances the temporal coherence between frames. The proposed model is evaluated on the SHRI-VID dataset (collected according to the task requirements), the AU-AIR dataset, and the ImageNet-VID benchmark. Extensive ablation studies and comparisons on detection and identity-awareness capacities show the superiority of the proposed model. Finally, a set of actual testing is conducted on a space robot, and the results show that the proposed model achieves a real-time speed and high accuracy.

Attention module improves both performance and interpretability of four-dimensional functional magnetic resonance imaging decoding neural network.

Decoding brain cognitive states from neuroimaging signals is an important topic in neuroscience. In recent years, deep neural networks (DNNs) have been recruited for multiple brain state decoding and achieved good performance. However, the open question of how to interpret the DNN black box remains unanswered. Capitalizing on advances in machine learning, we integrated attention modules into brain decoders to facilitate an in‐depth interpretation of DNN channels. A four‐dimensional (4D) convolution operation was also included to extract temporo‐spatial interaction within the fMRI signal. The experiments showed that the proposed model obtains a very high accuracy (97.4%) and outperforms previous researches on the seven different task benchmarks from the Human Connectome Project (HCP) dataset. The visualization analysis further illustrated the hierarchical emergence of task‐specific masks with depth. Finally, the model was retrained to regress individual traits within the HCP and to classify viewing images from the BOLD5000 dataset, respectively. Transfer learning also achieves good performance. Further visualization analysis shows that, after transfer learning, low‐level attention masks remained similar to the source domain, whereas high‐level attention masks changed adaptively. In conclusion, the proposed 4D model with attention module performed well and facilitated interpretation of DNNs, which is helpful for subsequent research.

CARNet: Cascade attentive RefineNet for multi-lesion segmentation of diabetic retinopathy images

Diabetic retinopathy is the leading cause of blindness in working population. Lesion segmentation from fundus images helps ophthalmologists accurately diagnose and grade of diabetic retinopathy. However, the task of lesion segmentation is full of challenges due to the complex structure, the various sizes and the interclass similarity with other fundus tissues. To address the issue, this paper proposes a cascade attentive RefineNet (CARNet) for automatic and accurate multi-lesion segmentation of diabetic retinopathy. It can make full use of the fine local details and coarse global information from the fundus image. CARNet is composed of global image encoder, local image encoder and attention refinement decoder. We take the whole image and the patch image as the dual input, and feed them to ResNet50 and ResNet101, respectively, for downsampling to extract lesion features. The high-level refinement decoder uses dual attention mechanism to integrate the same-level features in the two encoders with the output of the low-level attention refinement module for multiscale information fusion, which focus the model on the lesion area to generate accurate predictions. We evaluated the segmentation performance of the proposed CARNet on the IDRiD, E-ophtha and DDR data sets. Extensive comparison experiments and ablation studies on various data sets demonstrate the proposed framework outperforms the state-of-the-art approaches and has better accuracy and robustness. It not only overcomes the interference of similar tissues and noises to achieve accurate multi-lesion segmentation, but also preserves the contour details and shape features of small lesions without overloading GPU memory usage.

Multi-Modal fusion with multi-level attention for Visual Dialog

Given an input image, Visual Dialog is introduced to answer a sequence of questions in the form of a dialog. To generate accurate answers for questions in the dialog, we need to consider all information of the dialog history, the question, and the image. However, existing methods usually directly utilized the high-level semantic information of the whole sentence for the dialog history and the question, while ignoring the low-level detailed information of words in the sentence. Similarly, the detailed region information of the image in low level is also required to be considered for question answering. Therefore, we propose a novel visual dialog method, which focuses on both high-level and low-level information of the dialog history, the question, and the image. In our approach, we introduce three low-level attention modules, the goal of which is to enhance the representation of words in the sentence of the dialog history and the question based on the word-to-word connection and enrich the region information of the image based on the region-to-region relation. Besides, we design three high-level attention modules to select important words in the sentence of the dialog history and the question as the supplement of the detailed information for semantic understanding, as well as to select relevant regions in the image to provide the targeted visual information for question answering. We evaluate the proposed approach on two datasets: VisDial v0.9 and VisDial v1.0. The experimental results demonstrate that utilizing both low-level and high-level information really enhances the representation of inputs.

Neural Machine Translation with Deep Attention.

Deepening neural models has been proven very successful in improving the model's capacity when solving complex learning tasks, such as the machine translation task. Previous efforts on deep neural machine translation mainly focus on the encoder and the decoder, while little on the attention mechanism. However, the attention mechanism is of vital importance to induce the translation correspondence between different languages where shallow neural networks are relatively insufficient, especially when the encoder and decoder are deep. In this paper, we propose a deep attention model (DeepAtt). Based on the low-level attention information, DeepAtt is capable of automatically determining what should be passed or suppressed from the corresponding encoder layer so as to make the distributed representation appropriate for high-level attention and translation. We conduct experiments on NIST Chinese-English, WMT English-German, and WMT English-French translation tasks, where, with five attention layers, DeepAtt yields very competitive performance against the state-of-the-art results. We empirically find that with an adequate increase of attention layers, DeepAtt tends to produce more accurate attention weights. An in-depth analysis on the translation of important context words further reveals that DeepAtt significantly improves the faithfulness of system translations.

No meditation-related changes in the auditory N1 during first-time meditation

Recent studies link meditation expertise with enhanced low-level attention, measured through auditory event-related potentials (ERPs). In this study, we tested the reliability and validity of a recent finding that the N1 ERP in first-time meditators is smaller during meditation than non-meditation – an effect not present in long-term meditators. In the first experiment, we replicated the finding in first-time meditators. In two subsequent experiments, we discovered that this finding was not due to stimulus-related instructions, but was explained by an effect of the order of conditions. Extended exposure to the same tones has been linked with N1 decrement in other studies, and may explain N1 decrement across our two conditions. We give examples of existing meditation and ERP studies that may include similar condition order effects. The role of condition order among first-time meditators in this study indicates the importance of counterbalancing meditation and non-mediation conditions in meditation studies that use event-related potentials.

Sensory Metrics of Neuromechanical Trust.

Today digital sources supply a historically unprecedented component of human sensorimotor data, the consumption of which is correlated with poorly understood maladies such as Internet addiction disorder and Internet gaming disorder. Because both natural and digital sensorimotor data share common mathematical descriptions, one can quantify our informational sensorimotor needs using the signal processing metrics of entropy, noise, dimensionality, continuity, latency, and bandwidth. Such metrics describe in neutral terms the informational diet human brains require to self-calibrate, allowing individuals to maintain trusting relationships. With these metrics, we define the trust humans experience using the mathematical language of computational models, that is, as a primitive statistical algorithm processing finely grained sensorimotor data from neuromechanical interaction. This definition of neuromechanical trust implies that artificial sensorimotor inputs and interactions that attract low-level attention through frequent discontinuities and enhanced coherence will decalibrate a brain's representation of its world over the long term by violating the implicit statistical contract for which self-calibration evolved. Our hypersimplified mathematical understanding of human sensorimotor processing as multiscale, continuous-time vibratory interaction allows equally broad-brush descriptions of failure modes and solutions.For example, we model addiction in general as the result of homeostatic regulation gone awry in novel environments (sign reversal) and digital dependency as a sub-case in which the decalibration caused by digital sensorimotor data spurs yet more consumption of them. We predict that institutions can use these sensorimotor metrics to quantify media richness to improve employee well-being; that dyads and family-size groups will bond and heal best through low-latency, high-resolution multisensory interaction such as shared meals and reciprocated touch; and that individuals can improve sensory and sociosensory resolution through deliberate sensory reintegration practices. We conclude that we humans are the victims of our own success, our hands so skilled they fill the world with captivating things, our eyes so innocent they follow eagerly.

Social Models Enhance Apes\u2019 Memory for Novel Events

Nonhuman primates are more likely to learn from the actions of a social model than a non-social “ghost display”, however the mechanism underlying this effect is still unknown. One possibility is that live models are more engaging, drawing increased attention to social stimuli. However, recent research with humans has suggested that live models fundamentally alter memory, not low-level attention. In the current study, we developed a novel eye-tracking paradigm to disentangle the influence of social context on attention and memory in apes. Tested in two conditions, zoo-housed apes (2 gorillas, 5 chimpanzees) were familiarized to videos of a human hand (social condition) and mechanical claw (non-social condition) constructing a three-block tower. During the memory test, subjects viewed side-by-side pictures of the previously-constructed block tower and a novel block tower. In accordance with looking-time paradigms, increased looking time to the novel block tower was used to measure event memory. Apes evidenced memory for the event featuring a social model, though not for the non-social condition. This effect was not dependent on attention differences to the videos. These findings provide the first evidence that, like humans, social stimuli increase nonhuman primates’ event memory, which may aid in information transmission via social learning.

Meditation and auditory attention: An ERP study of meditators and non-meditators

The findings of a study by Cahn and Polich (2009) suggests that there is an effect of a meditative state on three event-related potential (ERP) brain markers of “low-level” auditory attention (i.e., acoustic representations in sensory memory) in expert meditators: the N1, the P2, and the P3a. The current study built on these findings by examining trait and state effects of meditation on the passive auditory mismatch negativity (MMN), N1, and P2 ERPs. We found that the MMN was significantly larger in meditators than non-meditators regardless of whether they were meditating or not (a trait effect), and that N1 amplitude was significantly attenuated during meditation in non-meditators but not expert meditators (an interaction between trait and state). These outcomes suggest that low-level attention is superior in long-term meditators in general. In contrast, low-level attention is reduced in non-meditators when they are asked to meditate for the first time, possibly due to auditory fatigue or cognitive overload.

Low-level visual attention and its relation to joint attention in autism spectrum disorder

Visual attention is integral to social interaction and is a critical building block for development in other domains (e.g., language). Furthermore, atypical attention (especially joint attention) is one of the earliest markers of autism spectrum disorder (ASD). The current study assesses low-level visual attention and its relation to social attentional processing in youth with ASD and typically developing (TD) youth, aged 7 to 18 years. The findings indicate difficulty overriding incorrect attentional cues in ASD, particularly with non-social (arrow) cues relative to social (face) cues. The findings also show reduced competition in ASD from cues that remain on-screen. Furthermore, social attention, autism severity, and age were all predictors of competing cue processing. The results suggest that individuals with ASD may be biased towards speeded rather than accurate responding, and further, that reduced engagement with visual information may impede responses to visual attentional cues. Once attention is engaged, individuals with ASD appear to interpret directional cues as meaningful. These findings from a controlled, experimental paradigm were mirrored in results from an ecologically valid measure of social attention. Attentional difficulties may be exacerbated during the complex and dynamic experience of actual social interaction. Implications for intervention are discussed.

The Long-Term Effects of E-Advertising: The Influence of Internet Pop-ups Viewed at a Low Level of Attention in Implicit Memory

Faced with an abundance of advertising messages, Internet users allocate only minimal cognitive resources to advertising. What are the effects of pop-up ads for a new brand viewed at low-level attention, and then measured when the Internet users have forgotten having seen them? In the theoretical context of processing fluency and implicit memory, the experiment n = 398 studied the effects of repeated brief exposure to different types of content words/image in pop-up ads 7 days and 3 months after exposure. The results show the overall positive effects of the pop-ups, the superiority of the image over words for effects on attitude toward the brand and the purchase intentions; but the words produce more semantic effects than the image.

Against the View that Consciousness and Attention are Fully Dissociable.

In this paper, I will try to show that the idea that there can be consciousness without some form of attention, and high-level top-down attention without consciousness, originates from a failure to notice the varieties of forms that top-down attention and consciousness can assume. I will present evidence that: there are various forms of attention and consciousness; not all forms of attention produce the same kind of consciousness; not all forms of consciousness are produced by the same kind of attention; there can be low-level attention (or preliminary attention), whether of an endogenous or exogenous kind, without consciousness; attention cannot be considered the same thing as consciousness.

Vigilance, expectancy, and noise: Attention in second language lexical learning and memory

Talamas et al. (1999), Ferré et al. (2006) and Sunderman and Kroll (2006) exposed participants to first-language/second-language (L1/L2) pairs of words and asked them to decide whether the second word was the correct translation of the first. In the critical condition, the L2 word was either the translation of the L1 word ( man → hombre) or a form-relative of the translation ( man → hambre). Less fluent speakers showed higher recognition latencies in the form-relative condition than did more fluent speakers. This report explores whether an appropriately trained Adaptive Resonance Theory (ART) neural network (Carpenter and Grossberg, 1987a) will suffer from form-relative interference, and the role of vigilance (a parameter of low-level attention sensitive to environmental complexity) in this effect. I argue that the learning environment of early bilinguals is more complex than that of adult L2 learners, and therefore adult learners may be less vigilant to word form. ART2 networks were trained with English and Spanish corpora under conditions emulating early and late second language acquisition at different vigilance levels and then serially exposed to the same types of word pairs used in the three studies mentioned above. Form-relative interference was observed, indicating that low-level attentional mechanisms may play a role in second language lexical learning and access.

Communication neuroscience as a tool for health psychologists.

Comments on the original article, "Increased attention but more efficient disengagement: Neuroscientific evidence for defensive processing of threatening health information" by L. T. E. Kessels, R. A. C. Ruiter, and B. M. Jansma (see record 2010-14873-005). Kessler et al present an example of Communication Neuroscience as a tool for understanding the mechanisms that lead some health messages to be processed in a way that facilitates impact whereas other messages are ignored. Kessels et al used event-related brain potentials (ERPs), a high temporal resolution method, to monitor neural activity in the moment that messages are presented. They use this technology to provide insight about the low-level attention processes through which individuals at highest risk (in this case, smokers) disengage from self-relevant health messages (threatening and nonthreatening smoking images). The findings of Kessels et al are convergent with prior theory and empirical work demonstrating that high threat messages may not achieve the desired effect if presented in isolation (Brown & Locker, 2009; Leventhal, Safer, & Panagis, 1983; Liberman & Chaiken, 1992). By using the tools of neuroscience, however, Kessels et al elucidate a mechanism that was not apparent through self-report or implicit (reaction time) measures. They demonstrate that high threat messages lead to increased attention capture, but more efficient disengagement when threatening messages are self-relevant; this in turn helps to explain why high threat messages may not have the desired effect, despite successfully capturing people's attention. As such, this study illustrates one benefit of combining the tools of neuroscience with more familiar methods in health psychology.

Increased attention but more efficient disengagement: Neuroscientific evidence for defensive processing of threatening health information.

Previous studies indicate that people respond defensively to threatening health information, especially when the information challenges self-relevant goals. The authors investigated whether reduced acceptance of self-relevant health risk information is already visible in early attention processes, that is, attention disengagement processes. In a randomized, controlled trial with 29 smoking and nonsmoking students, a variant of Posner's cueing task was used in combination with the high-temporal resolution method of event-related brain potentials (ERPs). Reaction times and P300 ERP. Smokers showed lower P300 amplitudes in response to high- as opposed to low-threat invalid trials when moving their attention to a target in the opposite visual field, indicating more efficient attention disengagement processes. Furthermore, both smokers and nonsmokers showed increased P300 amplitudes in response to the presentation of high- as opposed to low-threat valid trials, indicating threat-induced attention-capturing processes. Reaction time measures did not support the ERP data, indicating that the ERP measure can be extremely informative to measure low-level attention biases in health communication. The findings provide the first neuroscientific support for the hypothesis that threatening health information causes more efficient disengagement among those for whom the health threat is self-relevant.

Controlling gaze with an embodied interactive control architecture

Human-Robot Interaction (HRI) is a growing field of research that targets the development of robots which are easy to operate, more engaging and more entertaining. Natural human-like behavior is considered by many researchers as an important target of HRI. Research in Human-Human communications revealed that gaze control is one of the major interactive behaviors used by humans in close encounters. Human-like gaze control is then one of the important behaviors that a robot should have in order to provide natural interactions with human partners. To develop human-like natural gaze control that can integrate easily with other behaviors of the robot, a flexible robotic architecture is needed. Most robotic architectures available were developed with autonomous robots in mind. Although robots developed for HRI are usually autonomous, their autonomy is combined with interactivity, which adds more challenges on the design of the robotic architectures supporting them. This paper reports the development and evaluation of two gaze controllers using a new cross-platform robotic architecture for HRI applications called EICA (The Embodied Interactive Control Architecture), that was designed to meet those challenges emphasizing how low level attention focusing and action integration are implemented. Evaluation of the gaze controllers revealed human-like behavior in terms of mutual attention, gaze toward partner, and mutual gaze. The paper also reports a novel Floating Point Genetic Algorithm (FPGA) for learning the parameters of various processes of the gaze controller.

Impact of Tasks and Users' Characteristics on Virtual Reality Performance

A better understanding of how users perform virtual reality (VR) tasks may help build better VR interfaces. In this study, we concentrated on the compensatory behavior in VR depending on the tasks and users' characteristics. The tasks characteristics considered were display size (large display vs. desktop monitor) and tasks types (manipulation and travel). The users' characteristics studied were the visual attention abilities and users' satisfaction. Ninety-five subjects participated in the experimentation composed of two parts: the first one consisted in cognitive tests used to evaluate visual attention abilities, and the second one was based on a set of VR tasks. Our result showed that large displays positively affect on performance for some kinds of VR tasks. Moreover, this impact was linked to users' satisfaction and visual attention abilities. Indeed, users with low-level attention abilities and users who preferred the large display took more advantage of large displays. We concluded that large displays can be considered cognitive aids depending on the tasks and users' characteristics.

Selective attention-based novelty scene detection in dynamic environments

We propose a biologically motivated novelty detection model of a scene that can give a robust performance for natural color scenes with an affine transformed field of view, as well as noisy scenes in a dynamic visual environment. Novelty detection is an essential property for developmental robots. A topology of a visual scan path of an input scene and an energy signature for the corresponding visual scan path are obtained and considered when deciding on a novelty occurrence in an input scene. The visual scan path is generated by a low-level top-down attention model in conjunction with a bottom-up saliency map model.