Vector Semiotic Model for Visual Question Answering

In this paper, we propose a Vector Semiotic Model as a possible solution to the symbol grounding problem in the context of Visual Question Answering. The Vector Semiotic Model combines the advantages of a Semiotic Approach implemented in the Sign-Based World Model and Vector Symbolic Architectures. The Sign-Based World Model represents information about a scene depicted on an input image in a structured way and grounds abstract objects in an agent’s sensory input. We use the Vector Symbolic Architecture to represent the elements of the Sign-Based World Model on a computational level. Properties of a high-dimensional space and operations defined for high-dimensional vectors allow encoding the whole scene into a high-dimensional vector with the preservation of the structure. That leads to the ability to apply explainable reasoning to answer an input question. We conducted experiments are on a CLEVR dataset and show results comparable to the state of the art. The proposed combination of approaches, first, leads to the possible solution of the symbol-grounding problem and, second, allows expanding current results to other intelligent tasks (collaborative robotics, embodied intellectual assistance, etc.).     

Skilled players’ and novices’ difficulty anticipating left- vs. right-handed opponents’ action intentions varies across different points in time

A left-handers’ performance advantage in interactive sports is assumed to result from their relative rarity compared to right-handers. Part of this advantage may be explained by athletes facing difficulties anticipating left-handers’ action intentions, particularly when anticipation is based on kinematic cues available at an early stage of an opponent’s movement. Here we tested whether the type of volleyball attack is predicted better against right- vs. left-handed opponents’ movements and whether such handedness effects are evident at earlier time points in skilled players than novices. In a video-based experiment volleyball players and novices predicted the type of shot (i.e., smash vs. lob) of left- and right-handed volleyball attacks occluded at six different time points. Overall, right-handed attacks were better anticipated than left-handed attacks, volleyball players outperformed novices, and performance improved in later occlusion conditions. Moreover, in skilled players the handedness effect was most pronounced when attacks were occluded 480 ms prior to hand-ball-contact, whereas in novices it was most evident 240 ms prior to hand-ball-contact. Our findings provide further evidence of the effect of an opponent’s handedness on action outcome anticipation and suggest that its occurrence in the course of an opponent’s unfolding action likely depends on an observers’ domain-specific skill.     

Age-related effects of attentional and oculomotor capture by onsets and color singletons as a function of experience

The present experiment examined the degree to which experience with different stimulus characteristics affects attentional capture, particularly as related to aging. Participants were presented with onset target/color singleton distractor or color singleton target/onset distractor pairs across three experimental sessions. The target/distractor pairs were reversed in the second session such that the target in the first session became the distractor in the second and third sessions. For both young and old adults previous experience with color as a target defining feature influenced oculomotor capture with task-irrelevant color distractors. Experience with sudden onsets had the same effect for younger and older adults, although capture effects were substantially larger for onset than for color distractors. Experience-based capture effects diminished relatively rapidly after target and distractor-defining properties were reversed. The results are discussed in terms of top-down and stimulus-driven effects on age-related differences in attentional control.     

Infants’ scanning of dynamic faces during the first year

This research investigated infants’ scanning of a talking, socially engaging face. Three- to four-month-olds looked equally at the mouth and eyes whereas 9-month-olds attended more to the eyes than mouth. These findings shed light on information infants’ seek from dynamic face stimuli.     

Manipulation of visual information does not change the accuracy of distance estimation during a blindfolded walking task

While humans rely on vision during navigation, they are also competent at navigating non-visually. However, non-visual navigation over large distances is not very accurate and can accumulate error. Currently, it is unclear whether this accumulation of error is due to the visual estimate of the distance or to the locomotor production of the distance. In a series of experiments, using a blindfolded walking test, we examine whether enhancing the visual estimate of the distance to a previously seen target, through environmental enrichment, visual imagery, or repeated exposure would improve the accuracy of blindfold navigation across different distances. We also attempt to decrease the visual estimate in order to see if the opposite effect would occur. Our results would indicate that manipulation of the static visual distance estimate did not change the navigation accuracy to any great extent. The only condition that improved accuracy was repeated exposure to the environment through practice. These results suggest that error observed during blindfold navigation may be due to the locomotor production of the distance, rather than the visual process.     

Study of γ/γ′ interfacial width in a nickel-based superalloy by scanning transmission electron microscopy

The γ/γ′ interfaces are thought to play an important role in determining the mechanical properties in single-crystal nickel-based superalloys. In this article, interfacial width in DD6, one second-generation single-crystal superalloy containing 2 wt% Re, has been studied by means of scanning transmission electron microscopy. From an atomic resolution high angle annular dark field image, both compositional interfaces and order–disorder interfaces are studied, and it is first found that the interfacial width of chemical composition is same with that between the ordered and disordered phases.     

The influence of working memory on the anger superiority effect

The anger superiority effect shows that an angry face is detected more efficiently than a happy face. However, it is still controversial whether attentional allocation to angry faces is a bottom-up process or not. We investigated whether the anger superiority effect is influenced by top-down control, especially working memory (WM). Participants remembered a colour and then searched for differently coloured facial expressions. Just holding the colour information in WM did not modulate the anger superiority effect. However, when increasing the probabilities of trials in which the colour of a target face matched the colour held in WM, participants were inclined to direct attention to the target face regardless of the facial expression. Moreover, the knowledge of high probability of valid trials eliminated the anger superiority effect. These results suggest that the anger superiority effect is modulated by top-down effects of WM, the probability of events and expectancy about these probabilities.     

Better safe than sorry: Simplistic fear-relevant stimuli capture attention

It has been consistently demonstrated that fear-relevant images capture attention preferentially over fear-irrelevant images. Current theory suggests that this faster processing could be mediated by an evolved module that allows certain stimulus features to attract attention automatically, prior to the detailed processing of the image. The present research investigated whether simplified images of fear-relevant stimuli would produce interference with target detection in a visual search task. In Experiment 1, silhouettes and degraded silhouettes of fear-relevant animals produced more interference than did the fear-irrelevant images. Experiment 2, compared the effects of fear-relevant and fear-irrelevant distracters and confirmed that the interference produced by fear-relevant distracters was not an effect of novelty. Experiment 3 suggested that fear-relevant stimuli produced interference regardless of whether participants were instructed as to the content of the images. The three experiments indicate that even very simplistic images of fear-relevant animals can divert attention.     

Visual attention to emotion in depression: Facilitation and withdrawal processes

Attentional biases for sadness are integral to cognitive theories of depression, but do not emerge under all conditions. Some researchers have argued that depression is associated with delayed withdrawal from, but not facilitated initial allocation of attention toward, sadness. We compared two types of withdrawal processes in clinically depressed and non-depressed individuals: (1) withdrawal requiring overt eye movements during visual search; and (2) covert disengagement of attention in a modified cueing paradigm. We also examined initial allocation of attention towards emotion on the visual search task, allowing comparison of withdrawal and facilitation processes. As predicted, we found no evidence of facilitated attention towards sadness in depressed individuals. However, we also found no evidence of depression-linked differences in withdrawal of attention from sadness on either task, offering no support for the theory that depression is associated with withdrawal rather than initial facilitation of attention.     

fMRI evidence for dorsal stream processing abnormality in adults born preterm

We investigated the consequences of premature birth on the functional neuroanatomy of the dorsal stream of visual processing. fMRI was recorded while sixteen healthy participants, 8 (two men) adults (19 years 6 months old, SD 10 months) born premature (mean gestational age 30 weeks), referred to as Premas, and 8 (two men) matched controls (20 years 1 month old, SD 13 months), performed a 1-back memory task of Object or Grip information using a hand grasping a drinking vessel as stimulus. While history of prematurity did not significantly affect task performance, Group by Task analysis of variance in regions of interest spanning the occipital, temporal and parietal lobes revealed main effects of Task and interactions between the two factors. Object processing activated the left inferior occipital cortex and bilateral ventral temporal regions, belonging to the ventral stream, with no effect of Group. Grip processing across groups activated the early visual cortex and the left supramarginal gyrus belonging to the dorsal stream. Group effect on the brain activity during Grip suggested that Controls represented the actions’ goal while Premas relied more on low-level visual information. This shift from higher- to lower-order visual processing between Controls and Premas may reflect a more general trend, in which Premas inadequately recruit higher-order visual functions for dorsal stream task performance, and rely more on lower-level functions.