Yael Adini

Face recognition: the problem of compensating for changes in illumination direction

A face recognition system must recognize a face from a novel image despite the variations between images of the same face. A common approach to overcoming image variations because of changes in the illumination conditions is to use image representations that are relatively insensitive to these variations. Examples of such representations are edge maps, image intensity derivatives, and images convolved with 2D Gabor-like filters. Here we present an empirical study that evaluates the sensitivity of these representations to changes in illumination, as well as viewpoint and facial expression. Our findings indicated that none of the representations considered is sufficient by itself to overcome image variations because of a change in the direction of illumination. Similar results were obtained for changes due to viewpoint and expression. Image representations that emphasized the horizontal features were found to be less sensitive to changes in the direction of illumination. However, systems based only on such representations failed to recognize up to 20 percent of the faces in our database. Humans performed considerably better under the same conditions. We discuss possible reasons for this superiority and alternative methods for overcoming illumination effects in recognition.

Context-enabled learning in the human visual system

Training was found to improve the performance of humans on a variety of visual perceptual tasks. However, the ability to detect small changes in the contrast of simple visual stimuli could not be improved by repetition. Here we show that the performance of this basic task could be modified after the discrimination of the stimulus contrast was practised in the presence of similar laterally placed stimuli, suggesting a change in the local neuronal circuit involved in the task. On the basis of a combination of hebbian and anti-hebbian synaptic learning rules compatible with our results, we propose a mechanism of plasticity in the visual cortex that is enabled by a change in the context.

Perceptual learning in contrast discrimination: the effect of contrast uncertainty.

Performance in perceptual tasks improves with repetition (perceptual learning), eventually reaching a saturation level. Typically, when perceptual learning effects are studied, stimulus parameters are kept constant throughout the training and during the pre- and post-training tests. Here we investigate whether learning by repetition transfers to testing conditions in which the practiced stimuli are randomly interleaved during the post-training session. We studied practice effects with a contrast discrimination task, employing a number of training methods: (i) practice with a single, fixed pedestal (base-contrast), (ii) practice with several pedestals, and (iii) practice with several pedestals that included a spatial context. Pre- and post-training tests were carried out with the base contrast randomized across trials, under conditions of contrast uncertainty. The results showed that learning had taken place with the fixed pedestal method (i) and with the context method (iii), but only the latter survived the uncertainty test. In addition, we were able to identify a very fast learning phase in contrast discrimination that improved performance under uncertainty. We contend that learned tasks that do not pass the uncertainty test involve modification of decision strategies that require exact knowledge of the stimulus.

Abnormal Speech Spectrum and Increased Pitch Variability in Young Autistic Children

Children with autism spectrum disorder (ASD) who can speak often exhibit abnormal voice quality and speech prosody, but the exact nature and underlying mechanisms of these abnormalities, as well as their diagnostic power are currently unknown. Here we quantified speech abnormalities in terms of the properties of the long-term average spectrum (LTAS) and pitch variability in speech samples of 83 children (41 with ASD, 42 controls) ages 4–6.5 years, recorded while they named a sequence of daily life pictures for 60 s. We found a significant difference in the groups average spectra, with ASD spectra being shallower and exhibiting less harmonic structure. Contrary to the common impression of monotonic speech in autism, the ASD children had a significantly larger pitch range and variability across time. A measure of this variability, optimally tuned for the sample, yielded 86% success (90% specificity, 80% sensitivity) in classifying ASD in the sample. These results indicate that speech abnormalities in ASD are reflected in its spectral content and pitch variability. This variability could imply abnormal processing of auditory feedback or elevated noise and instability in the mechanisms that control pitch. The current results are a first step toward developing speech spectrum-based bio-markers for early diagnosis of ASD.

Recurrent networks in human visual cortex: psychophysical evidence

To study the neuronal circuitry underlying visual spatial-integration processes, we measured the effect of short and long chains of proximal Gabor-signal (GS) flankers (sigma = lambda = 0.15 degrees) on the contrast-discrimination function of a foveal GS target. We found that the same pattern of lateral masks enhanced target detection with low-contrast pedestals and strongly suppressed the discrimination of a range of intermediate pedestal contrasts (pedestal contrast <30%). Increasing the number of the flankers reversed the suppressive effect. The data suggest that the main influence of the proximal flankers is maintained by activity-dependent interactions and not by linear spatial summation. With an increased number of flankers, we found a nonmonotonic relationship between the discrimination thresholds and the number of flankers, supporting the notion that the discrimination thresholds are mediated by excitatory-inhibitory recurrent networks that manifest the dynamics of large neuronal populations in the neocortex [Proc. Natl. Acad. Sci. USA 94, 10426 (1997)].

Associative learning in early vision

Sensory discriminations often improve with practice (perceptual learning). Recent results show that practice does not necessarily lead to the best possible performance on the task. It was shown that learning a task (contrast discrimination) that has already reached saturation could be enabled by a contextual change in the stimulus (the addition of surrounding flankers) during practice. Psychophysical results with varying context show a behavior that is described by a network of local visual processors with horizontal recurrent interactions. We describe a mathematical learning rule for the modification of cortical synapses that is inspired by the experimental results and apply it to recurrent cortical networks that respond to external stimuli. The model predicts that repeated presentation of the same stimulus leads to saturation of synaptic modification, such that the strengths of recurrent connections depend on the configuration of the stimulus but not on its amplitude. When a new stimulus is introduced, the modification is rekindled until a new equilibrium is reached. This effect may explain the saturation of perceptual learning when practicing a certain task repeatedly. We present simulations of contrast discrimination in a simplified model of a cortical column in the primary visual cortex and show that performance of the model is reminiscent of context-dependent perceptual learning.

Parallel processes within the «spot-light» of attention

Human ability to identify simultaneously two targets in the visual field is severely limited. Previous studies have shown that orientation identification of two targets takes twice the time needed for one target. Here we asked whether this seriality is imposed by the decision requirement of the task or by such stimulus properties as target spatial separation and similarity. Observers had to identify the orientations (vertical vs horizontal) of two Gabor patches presented at random positions. Performance on this double-task experiment was compared with performance on each of the tasks when carried out alone. We varied the spatial separation between the two targets for targets having identical or different spatial-frequencies and found that the orientation of two targets having different frequencies could be identified in parallel when occupying the same spatial position but not when separated in space by 4 deg of visual angle or more. Targets having the same frequency could be identified in parallel even when separated by 8 deg, demonstrating that decision factors do not impose seriality. This result can be taken as evidence for the existence of a grouping process operating prior to orientation identification. This grouping process operates according to classical Gestalt rules (proximity, similarity) and enables parallel attentive processing of large input chunks.

Cross-modal extinction in a boy with severely autistic behaviour and high verbal intelligence.

Anecdotal reports from individuals with autism suggest a loss of awareness to stimuli from one modality in the presence of stimuli from another. Here we document such a case in a detailed study of A.M., a 13-year-old boy with autism in whom significant autistic behaviours are combined with an uneven IQ profile of superior verbal and low performance abilities. Although A.M.s speech is often unintelligible, and his behaviour is dominated by motor stereotypies and impulsivity, he can communicate by typing or pointing independently within a letter board. A series of experiments using simple and highly salient visual, auditory, and tactile stimuli demonstrated a hierarchy of cross-modal extinction, in which auditory information extinguished other modalities at various levels of processing. A.M. also showed deficits in shifting and sustaining attention. These results provide evidence for monochannel perception in autism and suggest a general pattern of winner-takes-all processing in which a stronger stimulus-driven representation dominates behaviour, extinguishing weaker representations.

The time course and characteristics of procedural learning in schizophrenia patients and healthy individuals

Patients with schizophrenia have deficits in some types of procedural learning. Several mechanisms contribute to this learning in healthy individuals, including statistical and sequence-learning. To find preserved and impaired learning mechanisms in schizophrenia, we studied the time course and characteristics of implicitly introduced sequence-learning (SRT task) in 15 schizophrenia patients (seven mild and eight severe) and nine healthy controls, in short sessions over multiple days (5–22). The data show speed gains of similar magnitude for all groups, but the groups differed in overall speed and in the characteristics of the learning. By analyzing the data according to its spatial-position and temporal-order components, we provide evidence for two types of learning that could differentiate the groups: while the learning of the slower, severe group was dominated by statistical learning, the control group moved from a fast learning phase of statistical-related performance to subsequence learning (chunking). Our findings oppose the naïve assumption that a similar gain of speed reflects a similar learning process; they indicate that the slower performance reflects the activation of a different motor plan than does the faster performance; and demonstrate that statistical learning and subsequence learning are two successive stages in implicit sequence learning, with chunks inferred from prior statistical computations. Our results indicate that statistical learning is intact in patients with schizophrenia, but is slower to develop in the severe patients. We suggest that this slow learning rate and the associated slow performance contribute to their deficit in developing sequence-specific learning by setting a temporal constraint on developing higher order associations.

Contrast sensitivity revealed by spontaneous eyeblinks: Evidence for a common mechanism of oculomotor inhibition.

Spontaneous eyeblinks are known to serve important physiological functions, and recent evidence shows that they are also linked to cognitive processes. It is yet unclear whether this link reflects a crude rate modulation or, alternatively, an automatic and precise process, tightly linked to the low-level properties of sensory stimuli. We have recently reported (Y. S. Bonneh, Adini, & Polat, 2015) that, for microsaccades, the onset and release from inhibition in response to transient stimuli depend systematically on the low-level stimulus parameters. Here we reanalyzed our previous data for both microsaccades and eyeblinks for observers with sufficient blinking (>10% of trials, 18 of 23 observers tested) who watched and silently counted sequences of Gabor patches at 1 Hz with varied contrast and spatial frequency. We found that spontaneous eyeblinks, although less frequent, were similar to microsaccades in their modulation pattern in response to transient stimuli, demonstrating inhibition and rebound, which were dependent on the contrast and spatial frequency of the stimuli. The average blink response time, measured as the latency of the first blink following its release from inhibition, was longer for lower contrast and higher spatial frequency. Importantly, it was highly correlated with a similar measure for microsaccades as well as with psychophysical measures of contrast sensitivity. These results suggest that both eyeblinks and microsaccades are linked to the same inhibitory mechanism that presumably turns off oculomotor events while processing previous events and generates a rebound effect upon its release. The onset of both eyeblinks and microsaccades may thus reflect the time course of this mechanism and the associated cognitive process.