Irrelevant Faces Research Articles

Taylor Swift does not boost face recognition in reaction time-based Concealed Information Test: investigating target-familiarity effects.

Eyewitness identifications from lineups are prone to error. More indirect identification procedures, such as the reaction-time based Concealed Information Test (RT-CIT) could be a viable alternative to lineups. The RT-CIT uses response times to assess facial familiarity. Theory and initial evidence with autobiographical stimuli suggests that the accuracy of RT-CIT can be augmented when participants' reliance on familiarity-based responding increases. We tested this assumption in two pre-registered experiments with 173 participants. Participants witnessed a mock crime. In the subsequent RT-CIT protocol, participants reacted to probe faces from the mock crime video, to irrelevant faces, and to target faces that required a unique response. Targets were either unknown people or were well-known celebrities (e.g., Taylor Swift). As expected, reaction times were longer to probes than to irrelevants in all conditions, indicating a CIT effect. Contrasting our pre-registered predictions, the CIT effect was not larger in the familiar condition (Experiment 1: unfamiliar targets: d = 0.77 vs. celebrity targets: d = 0.24; Experiment 2: unfamiliar targets: d = 1.09 vs. celebrity targets: d = 0.79). This suggests that familiar targets did not increase the validity of the RT-CIT in diagnosing concealed face recognition. A potential lack of saliency of the familiar targets might explain these unexpected findings. Of note, we did find medium to large effect sizes overall, speaking to the potential of diagnosing face recognition with the RT-CIT.

Disentangled deep generative models reveal coding principles of the human face processing network.

Despite decades of research, much is still unknown about the computations carried out in the human face processing network. Recently, deep networks have been proposed as a computational account of human visual processing, but while they provide a good match to neural data throughout visual cortex, they lack interpretability. We introduce a method for interpreting brain activity using a new class of deep generative models, disentangled representation learning models, which learn a low-dimensional latent space that "disentangles" different semantically meaningful dimensions of faces, such as rotation, lighting, or hairstyle, in an unsupervised manner by enforcing statistical independence between dimensions. We find that the majority of our model's learned latent dimensions are interpretable by human raters. Further, these latent dimensions serve as a good encoding model for human fMRI data. We next investigate the representation of different latent dimensions across face-selective voxels. We find that low- and high-level face features are represented in posterior and anterior face-selective regions, respectively, corroborating prior models of human face recognition. Interestingly, though, we find identity-relevant and irrelevant face features across the face processing network. Finally, we provide new insight into the few "entangled" (uninterpretable) dimensions in our model by showing that they match responses in the ventral stream and carry information about facial identity. Disentangled face encoding models provide an exciting alternative to standard "black box" deep learning approaches for modeling and interpreting human brain data.

Faster social attention disengagement in individuals with higher autism traits

ABSTRACT Introduction Atypical visual and social attention has often been associated with clinically diagnosed autism spectrum disorder (ASD), and with the broader autism phenotype. Atypical social attention is of particular research interest given the importance of facial expressions for social communication, with faces tending to attract and hold attention in neurotypical individuals. In autism, this is not necessarily so, where there is debate about the temporal differences in the ability to disengage attention from a face. Method Thus, we have used eye-tracking to record saccadic latencies as a measure of time to disengage attention from a central task-irrelevant face before orienting to a newly presented peripheral nonsocial target during a gap-overlap task. Neurotypical participants with higher or lower autism-like traits (AT) completed the task that included central stimuli with varied expressions of facial emotion as well as an inverted face. Results High AT participants demonstrated faster saccadic responses to detect the nonsocial target than low AT participants when disengaging attention from a face. Furthermore, faster saccadic responses were recorded when comparing disengagement from upright to inverted faces in low AT but not in high AT participants. Conclusions Together, these results extend findings of atypical social attention disengagement in autism and highlight how differences in attention to faces in the broader autism phenotype can lead to apparently superior task performance under certain conditions. Specifically, autism traits were linked to faster attention orienting to a nonsocial target due to the reduced attentional hold of the task irrelevant face stimuli. The absence of an inversion effect in high AT participants also reinforces the suggestion that they process upright or inverted faces similarly, unlike low AT participants for whom inverted faces are thought to be less socially engaging, thus allowing faster disengagement.

Task-irrelevant emotional faces impact BOLD responses more for prosaccades than antisaccades in a mixed saccade fMRI task

Cognitive control allows individuals to flexibly and efficiently perform tasks by attending to relevant stimuli while inhibiting distraction from irrelevant stimuli. The antisaccade task assesses cognitive control by requiring participants to inhibit a prepotent glance towards a peripheral stimulus and generate an eye movement to the mirror image location. This task can be administered with various contextual manipulations to investigate how factors such as trial timing or emotional content interact with cognitive control. In the current study, 26 healthy adults completed a mixed antisaccade and prosaccade fMRI task that included task irrelevant emotional faces and gap/overlap timing. The results showed typical antisaccade and gap behavioral effects with greater BOLD activation in frontal and parietal brain regions for antisaccade and overlap trials. Conversely, there were no differences in behavior based on the emotion of the task irrelevant face, but trials with neutral faces had greater activation in widespread visual regions than trials with angry faces, particularly for prosaccade and overlap trials. Together, these effects suggest that a high level of cognitive control and inhibition was required throughout the task, minimizing the impact of the face presentation on saccade behavior, but leading to increased attention to the neutral faces on overlap prosaccade trials when both the task cue (look towards) and emotion stimulus (neutral, non-threatening) facilitated disinhibition of visual processing.

Recognizing irrelevant faces in short-form videos based on feature fusion and active learning

In recent years, short-form videos spread rapidly around the world and became a popular way of entertainment for people to share their daily lives. However, many videos record behaviors of other people without their awareness and are uploaded onto the short-form video platforms. Such behavior severely invades personal privacy and can even bring risks of personal information leakage. At present, few studies focus on detecting privacy violations in short-form videos. Meanwhile, due to the difficulty in transferring existing models to the scenario of short-form videos and the lack of reliable datasets, it is very challenging to recognize irrelevant faces in short-form videos. To deal with this problem, we constructed and published an irrelevant faces dataset (IF-Dataset) with 43,965 irrelevant face images and 89,924 relevant face images based on the videos collected from Douyin (the Chinese version of TikTok). In addition, we constructed a framework that implemented our proposed deep learning model Multi-features Multi-head Fusion Network (MMFNet) to recognize irrelevant faces from short-form videos. The experimental results show that the F1 score of the MMFNet can reach 87.03%. We also proposed a novel loss function as well as an active learning system to improve the generalization ability of models, which can reach the Relative Error Reduction (RER) up to 29.58%. Our work provides both theoretical and practical support for face protection in short-form videos.

Probabilistic robust regression with adaptive weights — a case study on face recognition

Robust regression plays an important role in many machine learning problems. A primal approach relies on the use of Huber loss and an iteratively reweighted l2 method. However, because the Huber loss is not smooth and its corresponding distribution cannot be represented as a Gaussian scale mixture, such an approach is extremely difficult to handle using a probabilistic framework. To address those limitations, this paper proposes two novel losses and the corresponding probability functions. One is called Soft Huber, which is well suited for modeling non-Gaussian noise. Another is Nonconvex Huber, which can help produce much sparser results when imposed as a prior on regression vector. They can represent any lq loss ($${1 \over 2}$$ ⩽ q < 2) with tuning parameters, which makes the regression model more robust. We also show that both distributions have an elegant form, which is a Gaussian scale mixture with a generalized inverse Gaussian mixing density. This enables us to devise an expectation maximization (EM) algorithm for solving the regression model. We can obtain an adaptive weight through EM, which is very useful to remove noise data or irrelevant features in regression problems. We apply our model to the face recognition problem and show that it not only reduces the impact of noise pixels but also removes more irrelevant face images. Our experiments demonstrate the promising results on two datasets.

From individual features to full faces: Combining aspects of face information.

We investigated how information from face features is combined by comparing sensitivity to individual features with that for external (head shape, hairline) and internal (nose, mouth, eyes, eyebrows) feature compounds. Discrimination thresholds were measured for synthetic faces under the following conditions: (a) full-faces; (b) individual features (e.g., nose); and (c) feature compounds (either external or internal). Individual features and feature compounds were presented both in isolation and embedded within a fixed, task irrelevant face context. Relative to the full-face baseline, threshold elevations for the internal feature compound (2.41x) were comparable to those for the most sensitive individual feature (nose = 2.12x). External features demonstrated the same pattern. A model that incorporated all available feature information within a single channel in an efficient way overestimated sensitivity to feature compounds. Embedding individual features within a task-irrelevant context reduced discrimination sensitivity, relative to isolated presentation. Sensitivity to feature compounds, however, was unaffected by embedding. A loss of sensitivity when embedding features within a fixed-face context is consistent with holistic processing, which limits access to information about individual features. However, holistic combination of information across face features is not efficient: Sensitivity to feature compounds is no better than sensitivity to the best individual feature. No effect of embedding internal feature compounds within task-irrelevant external face features (or vice versa) suggests that external and internal features are processed independently.

Similar distraction, but differential suppression, for faces and non-face objects: Evidence from behaviour and event-related potentials

In everyday life we constantly experience distractions. Some distractors might be more distracting than others, for example the human face, which has been shown to be very efficient in attracting attention. Here, we employed the irrelevant-distractor paradigm (Forster & Lavie, 2016) to measure behavioural and neural distraction by completely irrelevant faces or non-faces (cars), while participants performed a letter search task that was more (high-load) or less (low-load) demanding. Under low load conditions, faces and cars equally slowed responses on trials with distractors as compared to those without. In high load conditions, neither faces nor cars were distracting. However, event-related potentials revealed larger face-sensitive N170 responses to faces than cars under both load conditions, suggesting that early face-specific processing is present even under high load. A subsequent Pd modulation between 200 and 300 ms contralateral to the distractor position, which has been linked to the active top-down suppression of lateral distractors (Hickey et al., 2009) was stronger for faces compared to cars. Overall, the EEG data indicate early face-specific processes to irrelevant faces irrespective of attentional load, coupled with stronger top-down suppression for faces. Together, these processes might reduce the influence of irrelevant face distractors on behavioural performance.

Examining a Novel Gamified Approach to Attentional Retraining: Effects of Single and Multiple Session Training

Negative attention biases (AB) may play a causal role in the development of emotional disorders. In order to examine this proposed causal role, researchers have developed Attention Bias Modification (ABM) paradigms to experimentally induce or reduce AB. To date, most ABM studies have been based on modified dot-probe tasks. However, this task is only moderately successful in changing patterns of AB. In two laboratory-based experiments, we explored the effects of a novel visual search ABM paradigm, called “Intrinsically-Motivating Playable Attentional Control Training”, on AB processes and mood in undergraduate participants. Motivation was fostered by active task involvement (i.e., searching for target faces while ignoring irrelevant faces) and gamification techniques. In both experiments, training performance significantly improved, but failed to transfer to attention and mood measures. Possible explanations for the failure to find transfer effects are discussed.

The power of a smile: Stronger working memory effects for happy faces in adolescents compared to adults

Theories of adolescent behaviour attribute increases in risk-taking and sensation-seeking in this age group to a heightened sensitivity to emotional stimuli on the one hand and a relatively immature cognitive control system on the other hand. However, little research has outlined to what extent relevant and irrelevant emotional stimuli bias the imbalance between affective processing and cognitive control. Thirty-three adolescents (19 females, aged 12–14) and 37 adults (18 females, aged 18–29) completed two attentional conditions of an emotional face working memory (WM) 0-back/2-back task. Participants were asked to attend to the emotional facial expression in the “relevant” emotion condition, and to the gender of the face in the “irrelevant” condition. The results revealed a WM improvement for happy faces in the relevant condition in both age groups, and an impairment for irrelevant happy faces in adolescents, but not adults. Furthermore, the difference between both attentional conditions for happy faces was larger in adolescents than adults. Results are discussed within the framework of theories of adolescent behaviour.

Culture modulates the brain response to human expressions of emotion: Electrophysiological evidence

To understand how culture modulates on-line neural responses to social information, this study compared how individuals from two distinct cultural groups, English-speaking North Americans and Chinese, process emotional meanings of multi-sensory stimuli as indexed by both behaviour (accuracy) and event-related potential (N400) measures. In an emotional Stroop-like task, participants were presented face–voice pairs expressing congruent or incongruent emotions in conditions where they judged the emotion of one modality while ignoring the other (face or voice focus task). Results indicated that while both groups were sensitive to emotional differences between channels (with lower accuracy and higher N400 amplitudes for incongruent face–voice pairs), there were marked group differences in how intruding facial or vocal cues affected accuracy and N400 amplitudes, with English participants showing greater interference from irrelevant faces than Chinese. Our data illuminate distinct biases in how adults from East Asian versus Western cultures process socio-emotional cues, supplying new evidence that cultural learning modulates not only behaviour, but the neurocognitive response to different features of multi-channel emotion expressions.

Attentional capture by completely task-irrelevant faces.

The present study investigated whether faces capture attention regardless of attentional set. The presentation of a face as a distractor during a visual search has been shown to impair performance relative to when the face was absent, implying that faces automatically attract attention. If attentional control is contingent on the observer's current goal, faces should not capture attention when they are irrelevant to the observer's attentional set. Previous studies demonstrating face-induced attentional capture used faces that were relevant to the task. Thus, a task in which faces were completely irrelevant to the observer's set was created. Participants identified a target letter among heterogeneously colored non-targets while ignoring a peripheral facial image that appeared as a brief distractor. No face-specific capture was observed when the target-distractor stimulus onset asynchrony (SOA) was long (Experiment 1). When the SOA was shortened, attentional capture by irrelevant faces was observed (Experiment 2). Experiment 3 extended this finding to all conditions, regardless of the attractiveness of faces. No such capture effect was found in Experiment 4 with inverted-face distractors. These results indicate that completely task-irrelevant faces break through top-down attentional set given a brief distractor-target SOA.

Are task irrelevant faces unintentionally processed? Implicit learning as a test case.

Faces are one of the most important signals for reading people's mental states. In sync with their apparent "chronic" (cross-situational) relevance, faces have been argued to be processed independently of the task one is currently performing. Many of these demonstrations have involved "capture of attention" or increased interference by faces functioning as distractors. Here we ask whether multiple repetitions of task irrelevant faces leave a trace in the system. Specifically, we tested whether repeating structures instantiated by task irrelevant faces are unintentionally or implicitly learned. Our findings indicate that although faces are indeed unique in that they are the only stimulus found to lead to implicit learning of complex rules when irrelevant, such learning is small in magnitude. Although these results support the conjecture that task irrelevant faces are processed, the functional significance of this learning needs to be assessed.

Prism adaptation does not alter configural processing of faces

Patients with hemispatial neglect (‘neglect’) following a brain lesion show difficulty responding or orienting to objects and events on the left side of space. Substantial evidence supports the use of a sensorimotor training technique called prism adaptation as a treatment for neglect. Reaching for visual targets viewed through prismatic lenses that induce a rightward shift in the visual image results in a leftward recalibration of reaching movements that is accompanied by a reduction of symptoms in patients with neglect. The understanding of prism adaptation has also been advanced through studies of healthy participants, in whom adaptation to leftward prismatic shifts results in temporary neglect-like performance. Interestingly, prism adaptation can also alter aspects of non-lateralised spatial attention. We previously demonstrated that prism adaptation alters the extent to which neglect patients and healthy participants process local features versus global configurations of visual stimuli. Since deficits in non-lateralised spatial attention are thought to contribute to the severity of neglect symptoms, it is possible that the effect of prism adaptation on these deficits contributes to its efficacy. This study examines the pervasiveness of the effects of prism adaptation on perception by examining the effect of prism adaptation on configural face processing using a composite face task. The composite face task is a persuasive demonstration of the automatic global-level processing of faces: the top and bottom halves of two familiar faces form a seemingly new, unknown face when viewed together. Participants identified the top or bottom halves of composite faces before and after prism adaptation. Sensorimotor adaptation was confirmed by significant pointing aftereffect, however there was no significant change in the extent to which the irrelevant face half interfered with processing. The results support the proposal that the therapeutic effects of prism adaptation are limited to dorsal stream processing.

Prism adaptation does not alter configural processing of faces.

Patients with hemispatial neglect ('neglect') following a brain lesion show difficulty responding or orienting to objects and events on the left side of space. Substantial evidence supports the use of a sensorimotor training technique called prism adaptation as a treatment for neglect. Reaching for visual targets viewed through prismatic lenses that induce a rightward shift in the visual image results in a leftward recalibration of reaching movements that is accompanied by a reduction of symptoms in patients with neglect. The understanding of prism adaptation has also been advanced through studies of healthy participants, in whom adaptation to leftward prismatic shifts results in temporary neglect-like performance. Interestingly, prism adaptation can also alter aspects of non-lateralised spatial attention. We previously demonstrated that prism adaptation alters the extent to which neglect patients and healthy participants process local features versus global configurations of visual stimuli. Since deficits in non-lateralised spatial attention are thought to contribute to the severity of neglect symptoms, it is possible that the effect of prism adaptation on these deficits contributes to its efficacy. This study examines the pervasiveness of the effects of prism adaptation on perception by examining the effect of prism adaptation on configural face processing using a composite face task. The composite face task is a persuasive demonstration of the automatic global-level processing of faces: the top and bottom halves of two familiar faces form a seemingly new, unknown face when viewed together. Participants identified the top or bottom halves of composite faces before and after prism adaptation. Sensorimotor adaptation was confirmed by significant pointing aftereffect, however there was no significant change in the extent to which the irrelevant face half interfered with processing. The results support the proposal that the therapeutic effects of prism adaptation are limited to dorsal stream processing.

Effects of emotional expression of an irrelevant face image on recognition of visual stimuli

Psychophysiological experiments were performed on 34 healthy subjects. We analyzed the accuracy and latency of motor response in recognizing two types of complex visual stimuli, animals and objects, which were presented immediately after a brief presentation of face images with different emotional expressions: anger, fear, happiness, and a neutral expression. We revealed the dependence of response latency on emotional expression of the masked face. The response latency was lower when the test stimuli were preceded by angry or fearful faces compared to happy or neutral faces. These effects depended on the type of stimulus and were more expressive when recognizing objects compared to animals. We found that the effects of emotional faces were related to personal features of the subjects that they exhibited in the emotional and communicative blocks of Cattell’s test and were more expressive in more sensitive, anxious, and pessimistic introverts. The mechanisms of the effects of unconsciously perceived emotional information on human visual behavior are discussed.

Intentionally forgetting other-race faces: Costs and benefits?

Eyewitnesses to events with multiple actors might be aware that during a subsequent investigation some actors will need to be remembered and others can be forgotten. Research on the directed-forgetting procedure suggests that when some information is cued to be forgotten, retention of other information is enhanced. In three experiments, directed-forgetting conditions were compared with control conditions to assess potential costs and benefits of forgetting other-race faces. In Experiment 1, undergraduate students (N = 148; mostly Caucasian) viewed all Black faces or all Asian faces followed by overt remember or forget cues. Participants in the directed-forgetting conditions of Experiments 2 and 3 received more covert cues instructing them to remember the faces of one race and to forget the faces of another race. In Experiment 2, undergraduate students (N = 116; all Caucasian) viewed Black and Asian faces within the context of a criminal storyline. In Experiment 3, undergraduate students (N = 94; all Caucasian) again viewed Black and Asian faces; however, the remember and forget cues were embedded in a noncriminal narrative. Although faces generally were forgotten on cue, forgetting some faces did not enhance memory for other faces. Furthermore, recognition of remember-cued faces was impaired by exposure to forget-cued faces. These findings indicate that faces can be forgotten on cue, but that doing so confers no benefit for remembering other faces. Eyewitnesses are advised that exposure to irrelevant faces reduces the likelihood that relevant faces will be remembered, even when effort is allocated to forgetting the irrelevant faces.

Irrelevant faces do not capture spatial attention in RSVP sequences

Irrelevant faces do not capture spatial attention in RSVP sequences

Oculomotor Guidance and Capture by Irrelevant Faces

Even though it is generally agreed that face stimuli constitute a special class of stimuli, which are treated preferentially by our visual system, it remains unclear whether faces can capture attention in a stimulus-driven manner. Moreover, there is a long-standing debate regarding the mechanism underlying the preferential bias of selecting faces. Some claim that faces constitute a set of special low-level features to which our visual system is tuned; others claim that the visual system is capable of extracting the meaning of faces very rapidly, driving attentional selection. Those debates continue because many studies contain methodological peculiarities and manipulations that prevent a definitive conclusion. Here, we present a new visual search task in which observers had to make a saccade to a uniquely colored circle while completely irrelevant objects were also present in the visual field. The results indicate that faces capture and guide the eyes more than other animated objects and that our visual system is not only tuned to the low-level features that make up a face but also to its meaning.

The presence of threat affects saccade trajectories

In everyday life, fast identification and processing of threat-related stimuli is of critical importance for survival. Previous studies suggested that spatial attention is automatically allocated to threatening stimuli, such as angry faces. However, in the previous studies the threatening stimuli were not completely irrelevant for the task. In the present study we used saccadic curvature to investigate whether attention is automatically allocated to threatening emotional information. Participants had to make an endogenous saccade up or down while an irrelevant face paired with an object was present in the periphery. The eyes curved away more from the angry faces than from either neutral or happy faces. This effect was not observed when the faces were inverted, excluding the possible role of low-level differences. Since the angry faces were completely irrelevant to the task, the results suggest that attention is automatically allocated to the threatening stimuli, which generates activity in the oculomotor system, and biases behaviour.