Feature-specific Information Research Articles

FEAT: A general framework for feature-aware multivariate time-series representation learning

Multivariate time-series is complex and uncertain. The overall temporal patterns change dynamically over time, and each feature is often observed to have a unique pattern. Therefore, it is challenging to model a framework that can flexibly learn feature-specific unique patterns as well as dynamically changing temporal patterns simultaneously. We propose a general framework for FEature-Aware multivariate Time-series representation learning, called FEAT. Unlike previous methods that only focus on training the overall temporal dependencies, we focus on training feature-specific as well as feature-agnostic representations in a data-driven manner. Specifically, we introduce a feature-wise encoder to explicitly model the feature-specific information and design an element-wise gating layer that learns the influence of feature-specific patterns per dataset in general. FEAT outperforms the benchmark models in average accuracy on 29 UEA multivariate time-series classification datasets and in MSE and MAE on four multivariate time-series forecasting datasets.

Jointly learning compact multi-view hash codes for few-shot FKP recognition

As a relatively new biometric trait, Finger-Knuckle-Print (FKP) plays a vital role in establishing a personal authentication system in modern society due to its rich discriminative features, low time cost in image capture and user-friendliness. However, most existing KFP descriptors are hand-crafted and fail to work well with limited training samples. In this paper, we propose a feature learning method for few-shot FKP recognition by jointly learning compact multi-view hash codes (JLCMHC) of a FKP image. We first form the multi-view data vectors (MVDV) to exploit the multiple feature-specific information from a FKP image. Then, we learn a feature projection to encode the MVDV into compact binary codes in an unsupervised manner, where 1) the variance of the learned feature codes on each view is maximized and 2) the difference of the inter-view binary codes is enlarged, so that the redundant information in MVDV is reduced and more informative features can be obtained. Lastly, we pool the binary codes into block-wise statistics features as the final descriptor for FKP representation and recognition. Experimental results on the existing benchmark FKP databases clearly show that the JLCMHC method outperforms the state-of-the-art FKP descriptors.

The Impact of Interruptions on Feature Change Detection within Dynamic Scenes

The goal of the present study was to examine how interruptions occurring in dynamic scenes affect the ability to detect perceptual changes during level 1 situation awareness (SA). Undergraduates were asked to watch 24 brief videos (half with interruptions) including 8 with perceptual feature changes. All videos were unique and contained multiple dynamic objects. Three different sets of instructions regarding the changes were given to successive groups: no information, limited information, and feature specific information. Of the eight changes, half occurred during a visual interruption and half with no interruption. Results showed that participants detected few changes, but detections increased when given more information about the nature of the changes in the absence of interruptions. The findings suggest that interruptions may facilitate the decay of an objects’ activation level in working memory and that level 1 SA may be particularly fragile when the visual scene is interrupted.

Automatic and feature-specific prediction-related neural activity in the human auditory system

Prior experience enables the formation of expectations of upcoming sensory events. However, in the auditory modality, it is not known whether prediction-related neural signals carry feature-specific information. Here, using magnetoencephalography (MEG), we examined whether predictions of future auditory stimuli carry tonotopic specific information. Participants passively listened to sound sequences of four carrier frequencies (tones) with a fixed presentation rate, ensuring strong temporal expectations of when the next stimulus would occur. Expectation of which frequency would occur was parametrically modulated across the sequences, and sounds were occasionally omitted. We show that increasing the regularity of the sequence boosts carrier-frequency-specific neural activity patterns during both the anticipatory and omission periods, indicating that prediction-related neural activity is indeed feature-specific. Our results illustrate that even without bottom-up input, auditory predictions can activate tonotopically specific templates.

Attention Enhances the Efficacy of Communication in V1 Local Circuits.

Attention is a critical component of visual perception; however, the mechanisms of attention at the granular level are poorly understood. One possible mechanism by which attention modulates neuronal activity is to control the efficacy of communication between connected neurons; however, it is unclear whether attention alters communication efficacy across a variety of neuronal circuits. In parallel, attentional modulation of neuronal firing rate is not uniform but depends upon the match between neuronal feature selectivity and the feature required for successful task completion. Here we tested whether modulation of communication efficacy is a viable mechanism of attention by assessing whether it is consistent across a variety of neuronal circuits and dependent upon the type of information conveyed in each circuit. We identified monosynaptically connected pairs of V1 neurons through cross-correlation of neuronal spike trains recorded in adult female macaque monkeys performing attention-demanding contrast-change detection tasks. Attention toward the stimulus in the receptive field of recorded neurons significantly facilitated the efficacy of communication among connected pairs of V1 neurons. The amount of attentional enhancement depended upon neuronal physiology, with larger facilitation for circuits conveying information about task-relevant features. Furthermore, presynaptic activity was more determinant of attentional enhancement of communication efficacy than postsynaptic activity, and feedforward local circuits often displayed the largest facilitation with attention. Together, these findings highlight attentional modulation of communication efficacy as a generalized mechanism of attention and demonstrate that attentional modulation at the granular level depends on the relevance of feature-specific information conveyed by neuronal circuits.SIGNIFICANCE STATEMENT How we pay attention to objects and locations in the visual environment has a profound impact on visual perception. Individual neurons in the visual cortex are similarly regulated by shifts in visual attention; however, the rules that govern whether and how attention alters neuronal activity are not known. In this study, we explored whether attention regulates communication between connected pairs of neurons in the primary visual cortex. We observed robust attentional facilitation of communication among these circuits. Furthermore, the extent to which the circuits were facilitated by attention depended on whether the information they conveyed was relevant for the particular attention task.

LRRTM1 underlies synaptic convergence in visual thalamus.

It has long been thought that the mammalian visual system is organized into parallel pathways, with incoming visual signals being parsed in the retina based on feature (e.g. color, contrast and motion) and then transmitted to the brain in unmixed, feature-specific channels. To faithfully convey feature-specific information from retina to cortex, thalamic relay cells must receive inputs from only a small number of functionally similar retinal ganglion cells. However, recent studies challenged this by revealing substantial levels of retinal convergence onto relay cells. Here, we sought to identify mechanisms responsible for the assembly of such convergence. Using an unbiased transcriptomics approach and targeted mutant mice, we discovered a critical role for the synaptic adhesion molecule Leucine Rich Repeat Transmembrane Neuronal 1 (LRRTM1) in the emergence of retinothalamic convergence. Importantly, LRRTM1 mutant mice display impairment in visual behaviors, suggesting a functional role of retinothalamic convergence in vision.

Saccadic remapping of object-selective information.

Saccadic remapping, a presaccadic increase in neural activity when a saccade is about to bring an object into a neuron's receptive field, may be crucial for our perception of a stable world. Studies of perception and saccadic remapping, like ours, focus on the presaccadic acquisition of information from the saccade target, with no direct reference to underlying physiology. While information is known to be acquired prior to a saccade, it is unclear whether object-selective or feature-specific information is remapped. To test this, we performed a series of psychophysical experiments in which we presented a peripheral, nonfoveated face as a presaccadic target. The target face disappeared at saccade onset. After making a saccade to the location of the peripheral target face (which was no longer visible), subjects misperceived the expression of a subsequent, foveally presented neutral face as being repelled away from the peripheral presaccadic face target. This effect was similar to a sequential shape contrast or negative aftereffect but required a saccade, because covert attention was not sufficient to generate the illusion. Additional experiments further revealed that inverting the faces disrupted the illusion, suggesting that presaccadic remapping is object-selective and not based on low-level features. Our results demonstrate that saccadic remapping can be an object-selective process, spatially tuned to the target of the saccade and distinct from covert attention in the absence of a saccade.

Memory for shape reactivates the lateral occipital complex

Memory is thought to be a constructive process in which the cortical regions associated with processing event features are reactivated during retrieval. Although there is evidence for non-detailed cortical reactivation during retrieval (e.g., memory for visual or auditory information reactivates the visual or auditory processing regions, respectively), there is limited evidence that memory can reactivate cortical regions associated with processing detailed, feature-specific information. Such evidence is critical to our understanding of the mechanisms of episodic retrieval. The present functional magnetic resonance imaging (fMRI) study assessed whether the lateral occipital complex (LOC), a region that preferentially processes shape, is associated with retrieval of shape information. During encoding, participants were presented with colored abstract shapes that were either intact or scrambled. During retrieval, colored disks were presented and participants indicated whether the corresponding shape was previously “intact” or “scrambled”. To assess whether conscious retrieval of intact shapes reactivated LOC, we conducted a conjunction of shape perception/encoding and accurate versus inaccurate retrieval of intact shapes, which produced many activations in LOC. To determine whether activity in LOC was specific to intact shapes, we conducted a conjunction of shape perception/encoding and intact versus scrambled shapes, which also produced many activations in LOC. Furthermore, memory for intact shapes in each hemifield produced contralateral activity in LOC (e.g., memory for left visual field intact shapes activated right LOC), which reflects the specific reinstatement of perception/encoding activity. The present results extend previous feature-specific memory reactivation evidence and support the view that memory is a constructive process.

Gamma Oscillations Underlie the Maintenance of Feature-Specific Information and the Contents of Visual Working Memory.

Visual working memory (VWM) sustains information online as integrated object representations. Neuronal mechanisms supporting the maintenance of feature-specific information have remained unidentified. Synchronized oscillations in the gamma band (30-120 Hz) characterize VWM retention and predict task performance, but whether these oscillations are specific to memorized features and VWM contents or underlie general executive VWM functions is not known. In the present study, we investigated whether gamma oscillations reflect the maintenance of feature-specific information in VWM. Concurrent magneto- and electroencephalography was recorded while subjects memorized different object features or feature conjunctions in identical VWM experiments. Using a data-driven source analysis approach, we show that the strength, load-dependence, and source topographies of gamma oscillations in the visual cortex differentiate these memorized features. Load-dependence of gamma oscillations in feature-specific visual and prefrontal areas also predicts VWM accuracy. Furthermore, corroborating the hypothesis that gamma oscillations support the perceptual binding of feature-specific neuronal assemblies, we also show that VWM for color-location conjunctions is associated with stronger gamma oscillations than that for these features separately. Gamma oscillations hence support the maintenance of feature-specific information and reflect VWM contents. The results also suggest that gamma oscillations contribute to feature binding in the formation of memory representations.

Saccadic remapping of object-selective information

Saccadic remapping, a presaccadic increase in neural activity when a saccade is about to bring an object into a neuron's receptive field, may be crucial for our perception of a stable world. Studies of perception and saccadic remapping, like ours, focus on the presaccadic acquisition of information from the saccade target, with no direct reference to underlying physiology. While information is known to be acquired prior to a saccade, it is unclear whether object-selective or feature-specific information is remapped. To test this, we performed a series of psychophysical experiments in which we presented a peripheral, nonfoveated face as a presaccadic target. The target face disappeared at saccade onset. After making a saccade to the location of the peripheral target face (which was no longer visible), subjects misperceived the expression of a subsequent, foveally presented neutral face as being repelled away from the peripheral presaccadic face target. This effect was similar to a sequential shape contrast or negative aftereffect but required a saccade, because covert attention was not sufficient to generate the illusion. Additional experiments further revealed that inverting the faces disrupted the illusion, suggesting that presaccadic remapping is object-selective and not based on low-level features. Our results demonstrate that saccadic remapping can be an object-selective process, spatially tuned to the target of the saccade and distinct from covert attention in the absence of a saccade.

The role of inner and outer face parts in holistic processing: A developmental study

The effects of inner–outer feature interactions with unfamiliar faces were investigated in 6- and 10-year-old children and adults (20–30years) to determine their contribution in holistic face vision. Participants completed a two-alternative forced-choice (2AFC) task under two conditions. The congruent condition used whole, inner-only, and outer-only stimuli. The incongruent condition used stimuli combining the inner features from one face with outer features from a novel face, or vice versa. Results yielded strong congruency effects which were moderated by pronounced feature-type asymmetries specific to developmental stage. Adults showed an inner-feature preference during congruent trials, but no asymmetry for incongruent trials. Children showed no asymmetry for congruent trials, but an outer-feature preference for incongruent trials. These findings concur with recent theoretical developments indicating that adults and children are likely to differ in the types of feature-specific information they preferentially encode in face perception, and that holistic effects are moderated differently in adults and children as a function of feature type.

Spatial specificity of working memory representations in the early visual cortex

Recent fMRI decoding studies have demonstrated that early retinotopic visual areas exhibit similar patterns of activity during the perception of a stimulus and during the maintenance of that stimulus in working memory. These findings provide support for the sensory recruitment hypothesis that the mechanisms underlying perception serve as a foundation for visual working memory. However, a recent study by Ester, Serences, and Awh (2009) found that the orientation of a peripheral grating maintained in working memory could be classified from both the contralateral and ipsilateral regions of the primary visual cortex (V1), implying that, unlike perception, feature-specific information was maintained in a nonretinotopic manner. Here, we evaluated the hypothesis that early visual areas can maintain information in a spatially specific manner and will do so if the task encourages the binding of feature information to a specific location. To encourage reliance on spatially specific memory, our experiment required observers to retain the orientations of two laterally presented gratings. Multivariate pattern analysis revealed that the orientation of each remembered grating was classified more accurately based on activity patterns in the contralateral than in the ipsilateral regions of V1 and V2. In contrast, higher extrastriate areas exhibited similar levels of performance across the two hemispheres. A time-resolved analysis further indicated that the retinotopic specificity of the working memory representation in V1 and V2 was maintained throughout the retention interval. Our results suggest that early visual areas provide a cortical basis for actively maintaining information about the features and locations of stimuli in visual working memory.

Feature specificity in attentional capture by size and color

Top-down guidance of visual attention has classically been thought to operate in a feature-specific manner. However, recent studies have shown that top-down visual attention can also be guided by information about target-nontarget feature relations (e.g., larger, redder, brighter). Here we recommend a minimal set of cues for differentiating between relational and feature-specific attentional guidance and examine contrasting predictions for the guidance of attention by size and color stimuli in a spatial cueing paradigm. In Experiment 1 we demonstrate that in search for size, when both feature-specific and relational strategies are available, participants adopt a relational search strategy. Experiment 2 shows that when feature-specific information is the only reliable information to guide attention to the target, participants are able to adopt a feature-specific set for size information. Finally, in Experiment 3 we extend our paradigm to differentiate between feature-specific and relational strategies in search for color. Together, these experiments help to clarify the conditions under which different attentional guidance strategies will be employed, and demonstrate a useful minimum cue requirement for differentiating between these two forms of top-down guidance. Implications for current theories of attention are discussed.

Oculomotor capture by colour singletons depends on intertrial priming

In visual search, an irrelevant colour singleton captures attention when the colour of the distractor changes across trials (e.g., from red to green), but not when the colour remains constant ( Becker, 2007). The present study shows that intertrial changes of the distractor colour also modulate oculomotor capture: an irrelevant colour singleton distractor was only selected more frequently than the inconspicuous nontargets (1) when its features had switched (compared to the previous trial), or (2) when the distractor had been presented at the same position as the target on the previous trial. These results throw doubt on the notion that colour distractors capture attention and the eyes because of their high feature contrast, which is available at an earlier point in time than information about specific feature values. Instead, attention and eye movements are apparently controlled by a system that operates on feature-specific information, and gauges the informativity of nominally irrelevant features.

Direct evidence for differential roles of temporal and frontal components of auditory change detection

Automatic change detection is a fundamental capacity of the human brain. In audition, this capacity is indexed by the mismatch negativity (MMN) event-related potential, which is putatively supported by a network consisting of superior temporal and frontal nodes. The aim of this study was to elucidate the roles of these nodes within the neural network of change detection. We used a dichotic paradigm in which subjects (N=14) attended targets defined by either pitch or spatial location in one auditory stream while the MMN was measured in response to unattended deviants of pitch and spatial location in the other stream. The frontal and temporal components of the MMN were examined using current source density (CSD) measurements. Competition for processing resources nearly eliminated the temporal CSD mismatch response, in a highly feature-specific manner: the response to spatial location deviants was reduced when the target dimension was spatial location but not when it was pitch, whereas the reverse occurred for pitch deviants. In contrast, the frontal CSD mismatch response was neither affected by competition nor by general attention demands. Thus, within the network of change detection, the temporal generators are specifically associated with processing feature-specific information, whereas the role of the frontal generators remains unclear. Moreover, the results are inconsistent with a serial model in which the frontal generator is contingent on activation of the temporal generator.

Feature versus gestalt representation of stimuli in the mismatch negativity system of 7‐ to 9‐year‐old children

We examined preattentive auditory change detection in 7- to 9-year-old children. The question of interest was whether the preattentive comparison of stimuli indexed by the scalp-recorded mismatch negativity (MMN) was performed on representations of individual stimulus features or on gestalt representations of their combined attributes. The design of the study, based on a work by D. Deacon, J. Nousak, M. Pilotti, W. Ritter, and C. Yang (Psychophysiology, 1998), was such that both feature and gestalt representations could have been available to the comparator mechanism generating the MMN. The data indicated that for the majority of the children-those that exhibited an inverse relationship between the amplitude of the MMN and the probability of the deviant-the MMN was based on feature-specific information. This study also provides a method to obtain MMNs to deviants in three different features in the time usually required to obtain an MMN to a single acoustic feature.

Inhibitory and excitatory responses of single neurons in the human medial temporal lobe during recognition of faces and objects.

The medial temporal lobe (MTL) plays a critical role in transforming complex stimuli into permanent memory traces, yet little is known on how the activity of neurons in the human brain mediates this process. Recording from single neurons in the human MTL during visual encoding and retrieval of faces and objects, we found that in the hippocampus faces evoked predominantly suppression of neuronal firing below prestimulus baseline ('inhibitory responses'). These responses were also prevalent in the entorhinal cortex but were absent in the amygdala during the first second of stimulus encoding when all responses to faces were 'excitatory' (neuronal firing increased above the prestimulus baseline). Inhibitory responses were more prevalent during recognition than encoding and were also present during processing of objects, albeit less frequently than during processing of faces. Despite the prevalence in the hippocampus of cells with inhibitory responses and their relative specificity to faces, it was mainly the activity of the cells with excitatory responses that was selective for stimulus features such as gender and emotional expression of faces. These findings suggest that a large population of cells with inhibitory responses is engaged in the hippocampal memory network, but primarily cells with excitatory responses process feature-specific information.

Hypervigilance in patients with obsessive-compulsive disorder.

The hypothesis that patients with obsessive-compulsive disorder (OCD) suffer from hypervigilant attention was investigated via explicit memory (incidental recall and recognition) and priming (reading speed) measures. OCD patients did not differ from normal controls on explicit measures of memory; specifically, recognition of unusual words (experiment 1) and recall and recognition of words and feature-specific information (experiment 2). Although both normal controls and OCD patients showed priming, the pattern of priming differed for the two groups (experiment 2). Specifically, patients with OCD failed to show feature-specific priming, suggesting they may have attended more focally on the priming task than did normal controls. These findings support previous reports of normal performance in OCD on explicit memory tasks, but suggest more sensitive measures may reflect differences in processing information.

Aging and feature-specific priming of familiar and novel stimuli.

English and Turkish words in 2 fonts were presented to young and elderly Ss, and memory was assessed directly with recognition tests and indirectly with reading time. Young and elderly Ss showed repetition priming on English words and showed larger priming effects when the study and test fonts were the same rather than different. Elderly Ss were impaired, however, on recognition of feature-specific information in comparison with young Ss. When Turkish words were shown, only young Ss evidenced priming, despite recognition performance above chance for both young and elderly Ss. A 2nd experiment replicated these findings with different materials.