ACE-GCN: A Fast Data-driven FPGA Accelerator for GCN Embedding

ACE-GCN is a fast and resource/energy-efficient FPGA accelerator for graph convolutional embedding under data-driven and in-place processing conditions. Our accelerator exploits the inherent power law distribution and high sparsity commonly exhibited by real-world graphs datasets. Contrary to other hardware implementations of GCN, on which traditional optimization techniques are employed to bypass the problem of dataset sparsity, our architecture is designed to take advantage of this very same situation. We propose and implement an innovative acceleration approach supported by our “implicit-processing-by-association” concept, in conjunction with a dataset-customized convolutional operator. The computational relief and consequential acceleration effect arise from the possibility of replacing rather complex convolutional operations for a faster embedding result estimation. Based on a computationally inexpensive and super-expedited similarity calculation, our accelerator is able to decide from the automatic embedding estimation or the unavoidable direct convolution operation. Evaluations demonstrate that our approach presents excellent applicability and competitive acceleration value. Depending on the dataset and efficiency level at the target, between 23× and 4,930× PyG baseline, coming close to AWB-GCN by 46% to 81% on smaller datasets and noticeable surpassing AWB-GCN for larger datasets and with controllable accuracy loss levels. We further demonstrate the unique hardware optimization characteristics of our approach and discuss its multi-processing potentiality.

The Long TBLT Journey

In this piece, I trace Task-Based Language Teaching from Mike Long’s original conceptualization in 1985, through his development of methodological principles that are based on SLA theory and empirical evidence he gathered for three and half decades, to two recent (2010, 2015) practical examples of TBLT. Since there is still some important work to be done on remaining tricky issues, I highlight a few of these (unresolved in instructed SLA), such as sequencing according to complexity and the resilience of implicit processing mechanisms in adult language learning. In resolving these and other issues that will no doubt arise, I urge us all to follow Mike’s scrupulous lead in holding ourselves accountable to the empirical evidence.

Story reading with incidental comprehension and memory: left hemisphere dominance

ABSTRACT Background: There are no studies on adults with unilateral brain lesions regarding story reading with incidental/implicit comprehension and memory, in which memory is only assessed through delayed recall. There is a need for validation of cerebral laterality in this type of verbal recall, which includes spontaneous performance (free or uncued condition (UC)), and induced-through-question performance regarding the forgotten units (cued condition (CC)). Objectives: To explore the effects of unilateral brain lesions, of oral reading with expression (RE) and comprehension (RC) on delayed recall of a story, as either UC or CC; and to validate the ability of UC and CC to discriminate the side of brain injury. Methods: Data were obtained from 200 right-handed volunteers, among whom 42 had left-hemisphere injury (LHI), 49 had right-hemisphere injury (RHI) and 109 were demographically-matched healthy participants (HP). Patients who were unable to read, understand or speak were excluded. Results: LHI individuals presented impairment of both UC and CC, in relation to the other two groups (non-LHI) with sensitivity and specificity above 70%. LHI and RHI individuals were not significantly different in RE and RC, but they were both different from HP in all the assessments except CC, in which RHI individuals resembled HP. Despite this lack of abnormality in RHI individuals during CC, about half of this group showed impairment in UC. Additionally, whereas RE had a significant effect on UC, the moral of the story (RC) had a significant effect on both UC and CC. Conclusions: The left hemisphere was dominant for this memory task involving implicit processing.

Danger or Dislike: Distinguishing Threat from Negative Valence as Sources of Automatic Anti-Black Bias

The Dual Implicit Process Model (March et al., 2018a) distinguishes the implicit processing of physical threat (i.e., “can it hurt or kill me?”) from valence (i.e., “do I dislike/like it?”). Five studies tested whether automatic anti-Black bias is due to White Americans associating Black men with threat, negative valence, or both. Studies 1 and 2 assessed how quickly White participants decided whether positive, negative, and threatening images were good versus bad when primed by Black versus White male-faces. Studies 3 and 4 assessed how early in the decision process White participants began deciding whether Black and White (and, in Study 3, Asian) male-faces displaying anger, sadness, happiness, or no emotion were, in Study 3, dangerous, depressed, cheerful, or calm or, in Study 4, dangerous, negative, or positive. Study 5 assessed how quickly White participants decided whether negative and threatening words were negative versus dangerous when primed by Black versus White male-names. All studies indicated that White Americans automatically associate Black men with physical threat. Study 3 indicated the association is unique to Black men and did not extend to Asian men as a general intergroup effect. Studies 3, 4, and 5, which simultaneously paired threat against negativity, indicated that the Black-threat association is stronger than a Black-negative association.

Implicit assumptions and interpretation bias in youth with severe, chronic social phobia

Interpretation bias and dysfunctional social assumptions are proposed to play a pivotal role in the development and maintenance of social phobia (SP), especially in youth. In this study, we aimed to investigate disorder-specific implicit assumptions of rejection and implicit interpretation bias in youth with severe, chronic SP and healthy controls (CG). Twenty-seven youth with SP in inpatient/day-care treatment (M age = 15.6 years, 74% female) and 24 healthy controls (M age = 15.7 years, 54% female) were included. The Implicit Association Test (IAT) and the Affect Misattribution Procedure (AMP) were completed to assess implicit assumptions and interpretation bias related to the processing of social and affective stimuli. No group differences were observed for the IAT controlling for depressive symptoms in the analyses. However, group differences were found regarding interpretation bias (p = .017, η2p = .137). Correlations between implicit scores and explicit questionnaire results were medium to large in the SP group (r =|.28| to |.54|, pall ≤ .05), but lower in the control group (r =|.04| to |.46|, pall ≤ .05). Our results confirm the finding of an interpretation bias in youth SP, especially regarding the implicit processing of faces, whereas implicit dysfunctional social assumptions of being rejected do not seem to be specific for SP. Future research should investigate the causal relationship of assumptions/interpretation bias and SP.

Energy optimization during walking involves implicit processing

ABSTRACT Gait adaptations, in response to novel environments, devices or changes to the body, can be driven by the continuous optimization of energy expenditure. However, whether energy optimization involves implicit processing (occurring automatically and with minimal cognitive attention), explicit processing (occurring consciously with an attention-demanding strategy) or both in combination remains unclear. Here, we used a dual-task paradigm to probe the contributions of implicit and explicit processes in energy optimization during walking. To create our primary energy optimization task, we used lower-limb exoskeletons to shift people's energetically optimal step frequency to frequencies lower than normally preferred. Our secondary task, designed to draw explicit attention from the optimization task, was an auditory tone discrimination task. We found that adding this secondary task did not prevent energy optimization during walking; participants in our dual-task experiment adapted their step frequency toward the optima by an amount and at a rate similar to participants in our previous single-task experiment. We also found that performance on the tone discrimination task did not worsen when participants were adapting toward energy optima; accuracy scores and reaction times remained unchanged when the exoskeleton altered the energy optimal gaits. Survey responses suggest that dual-task participants were largely unaware of the changes they made to their gait during adaptation, whereas single-task participants were more aware of their gait changes yet did not leverage this explicit awareness to improve gait adaptation. Collectively, our results suggest that energy optimization involves implicit processing, allowing attentional resources to be directed toward other cognitive and motor objectives during walking.

Impaired Sense of Agency and Associated Confidence in Psychosis

The Sense of Agency (SoA), our sensation of control over our actions, is a fundamental mechanism for delineating the Self from the environment and others. SoA arises from implicit processing of sensorimotor signals as well as explicit higher-level judgments. Psychosis patients suffer from difficulties in the sense of control over their actions and accurate demarcation of the Self. Moreover, it is unclear if they have metacognitive insight into their aberrant abilities. In this pre-registered study, we examined SoA and its associated confidence judgments using an embodied virtual reality paradigm in psychosis patients and controls. Our results show that psychosis patients not only have a severely reduced ability for discriminating their actions but they also do not show proper metacognitive insight into this deficit. Furthermore, an exploratory analysis revealed that the SoA capacities allow for high levels of accuracy in clinical classification of psychosis. These results indicate that SoA and its metacognition are core aspects of the psychotic state and provide possible venues for understanding the underlying mechanisms of psychosis, that may be leveraged for novel clinical purposes.

Preserved Perspective Taking in Free Indirect Discourse in Autism Spectrum Disorder

Perspective taking has been proposed to be impaired in persons with autism spectrum disorder (ASD), especially when implicit processing is required. In narrative texts, language perception and interpretation is fundamentally guided by taking the perspective of a narrator. We studied perspective taking in the linguistic domain of so-called Free Indirect Discourse (FID), during which certain text segments have to be interpreted as the thoughts or utterances of a protagonist without explicitly being marked as thought or speech representations of that protagonist (as in direct or indirect discourse). Crucially, the correct interpretation of text segments as FID depends on the ability to detect which of the protagonists “stands out” against the others and is therefore identifiable as implicit thinker or speaker. This so-called “prominence” status of a protagonist is based on linguistic properties (e.g., grammatical function, referential expression), in other words, the perspective is “hidden” and has to be inferred from the text material. In order to test whether this implicit perspective taking ability that is required for the interpretation of FID is preserved in persons with ASD, we presented short texts with three sentences to adults with and without ASD. In the last sentence, the perspective was switched either to the more or the less prominent of two protagonists. Participants were asked to rate the texts regarding their naturalness. Both diagnostic groups rated sentences with FID anchored to the less prominent protagonist as less natural than sentences with FID anchored to the more prominent protagonist. Our results that the high-level perspective taking ability in written language that is required for the interpretation of FID is well preserved in persons with ASD supports the conclusion that language skills are highly elaborated in ASD so that even the challenging attribution of utterances to protagonists is possible if they are only implicitly given. We discuss the implications in the context of claims of impaired perspective taking in ASD as well as with regard to the underlying processing of FID.

Case of Asperger's Syndrome and Lesion of the Right Amygdala: Deficits in Implicit and Explicit Fearful Face Recognition

Introduction: Studies of brain-damaged patients revealed that amygdala lesions cause deficits in the processing and recognition of emotional faces. Patients with autism spectrum disorders (ASD) have similar deficits also related to dysfunctions of the limbic system including the amygdala.Methods: We investigated a male patient who had been diagnosed with Asperger's syndrome. He also presented with a lesion of the right mesial temporal cortex, including the amygdala. We used functional magnetic resonance imaging (fMRI) to investigate neuronal processing during a passive viewing task of implicit and explicit emotional faces. Clinical assessment included a facial emotion recognition task.Results: There was no amygdala activation on both sides during the presentation of masked emotional faces compared to the no-face control condition. Presentation of unmasked happy and angry faces activated the left amygdala compared to the no-face control condition. There was no amygdala activation in response to unmasked fearful faces on both sides. In the facial emotion recognition task, the patient biased positive and neutral expressions as negative.Conclusions: This case report describes a male patient with right amygdala damage and an ASD. He displayed a non-response of the amygdala to fearful faces and tended to misinterpret fearful expressions. Moreover, a non-reactivity of both amygdalae to emotional facial expressions at an implicit processing level was revealed. It is discussed whether the deficient implicit processing of facial emotional information and abnormalities in fear processing could contribute and aggravate the patient's impairments in social behavior and interaction.

Right Hemisphere Dominance for Unconscious Emotionally Salient Stimuli

The present review will focus on evidence demonstrating the prioritization in visual processing of fear-related signals in the absence of awareness. Evidence in hemianopic patients without any form of blindsight or affective blindsight in classical terms will be presented, demonstrating that fearful faces, via a subcortical colliculo-pulvinar-amygdala pathway, have a privileged unconscious visual processing and facilitate responses towards visual stimuli in the intact visual field. Interestingly, this fear-specific implicit visual processing in hemianopics has only been observed after lesions to the visual cortices in the left hemisphere, while no effect was found in patients with damage to the right hemisphere. This suggests that the subcortical route for emotional processing in the right hemisphere might provide a pivotal contribution to the implicit processing of fear, in line with evidence showing enhanced right amygdala activity and increased connectivity in the right colliculo-pulvinar-amygdala pathway for unconscious fear-conditioned stimuli and subliminal fearful faces. These findings will be discussed within a theoretical framework that considers the amygdala as an integral component of a constant and continuous vigilance system, which is preferentially invoked with stimuli signaling ambiguous environmental situations of biological relevance, such as fearful faces.