Digit Categorization Research Articles

Suitability of V1 Energy Models for Object Classification

Simulations of cortical computation have often focused on networks built from simplified neuron models similar to rate models hypothesized for V1 simple cells. However, physiological research has revealed that even V1 simple cells have surprising complexity. Our computational simulations explore the effect of this complexity on the visual system's ability to solve simple tasks, such as the categorization of shapes and digits, after learning from a limited number of examples. We use recently proposed high-throughput methodology to explore what axes of modeling complexity are useful in these categorization tasks. We find that complex cell rate models learn to categorize objects better than simple cell models, and without incurring extra computational expense. We find that the squaring of linear filter responses leads to better performance. We find that several other components of physiologically derived models do not yield better performance.

Decoding cognitive control in the parietal cortex

We have previously reported evidence for a transient reconfiguration signal in the medial superior parietal lobule (mSPL/precuneus) that is time-locked to the initiation of shifts of attention within several perceptual domains, (e.g., attention shifts between spatial locations, features, objects or sensory modalities; Yantis et al., 2002; Liu et al., 2004; Serences et al., 2004, Shomstein & Yantis, 2004). Independently, studies of task-switching have revealed a frontoparietal cortical network exhibiting transient activity during switches between two categorization tasks (e.g., Braver et al., 2003). Here we investigated the neural basis of cognitive control during task-switching and spatial attention shifting within the same paradigm. Subjects directed covert attention to one of two rapid serial visual presentation (RSVP) streams displayed to the left and right of fixation, and were prepared to perform one of two digit categorization tasks (high/low or even/odd). Each stream contained mostly filler distractor items along with occasional cues and digit targets. Four types of cues instructed subjects to shift or hold spatial attention or to switch or hold their categorization task. Using the standard univariate GLM, we observed a transient reconfiguration signal evoked by both attention shift and task switch cues in a common region of mSPL/precuneus. We then applied multivoxel pattern classification (linear support vector machine) to determine whether this apparently domain-independent signal differs for the two domains of cognitive control. The multivoxel patterns of activity within this region did predict which type of shift was being executed on an individual trial basis. The multivariate analysis reveals that although this region is recruited for shifts in multiple domains, domain-specific spatial patterns of activity are associated with different acts of control.

Danger is worse when it moves: Neural and behavioral indices of enhanced attentional capture by dynamic threatening stimuli

Both dynamic non-emotional stimuli (moving dots or digits) and danger-related static stimuli have previously shown to capture attention. This study explored whether the combination of the two factors (i.e., threatening moving stimuli), frequent in natural situations, enhances attentional capture. To this end, static and moving distractors containing emotionally negative and non-negative information were presented to 30 volunteers while they were engaged in a digit categorization task. Behavioral responses and event-related potentials (ERPs) were analyzed. Behavioral and electrophysiological data were convergent: moving negative distractors produced the longest reaction times in the digit categorization task, and elicited the highest amplitudes in the P1 component of the ERPs (peaking at 112ms), an electrophysiological signal of attentional capture. These results suggest that motion provides additional salience to threatening stimuli that facilitates attentional capture.

Parametric Embedding for Class Visualization

We propose a new method, parametric embedding (PE), that embeds objects with the class structure into a low-dimensional visualization space. PE takes as input a set of class conditional probabilities for given data points and tries to preserve the structure in an embedding space by minimizing a sum of Kullback-Leibler divergences, under the assumption that samples are generated by a gaussian mixture with equal covariances in the embedding space. PE has many potential uses depending on the source of the input data, providing insight into the classifier's behavior in supervised, semisupervised, and unsupervised settings. The PE algorithm has a computational advantage over conventional embedding methods based on pairwise object relations since its complexity scales with the product of the number of objects and the number of classes. We demonstrate PE by visualizing supervised categorization of Web pages, semisupervised categorization of digits, and the relations of words and latent topics found by an unsupervised algorithm, latent Dirichlet allocation.

Beyond the data deluge: Data integration and bio-ontologies

Biomedical research is increasingly a data-driven science. New technologies support the generation of genome-scale data sets of sequences, sequence variants, transcripts, and proteins; genetic elements underpinning understanding of biomedicine and disease. Information systems designed to manage these data, and the functional insights (biological knowledge) that come from the analysis of these data, are critical to mining large, heterogeneous data sets for new biologically relevant patterns, to generating hypotheses for experimental validation, and ultimately, to building models of how biological systems work. Bio-ontologies have an essential role in supporting two key approaches to effective interpretation of genome-scale data sets: data integration and comparative genomics. To date, bio-ontologies such as the Gene Ontology have been used primarily in community genome databases as structured controlled terminologies and as data aggregators. In this paper we use the Gene Ontology (GO) and the Mouse Genome Informatics (MGI) database as use cases to illustrate the impact of bio-ontologies on data integration and for comparative genomics. Despite the profound impact ontologies are having on the digital categorization of biological knowledge, new biomedical research and the expanding and changing nature of biological information have limited the development of bio-ontologies to support dynamic reasoning for knowledge discovery.

Attention and expertise in multiple target tracking

AbstractAn experiment is reported examining the role of attention and expertise in multiple target tracking. It compares the ability of professional radar operators (experts) and undergraduate students (novices) to acquire and track subsets of randomly moving targets amongst distractors, and respond appropriately to a probe. Participants undertook this tracking task alone then, subsequently, in conjunction with a second, digit categorization task. For half the participants, the second task commenced prior to the tracking task's target acquisition phase, for the remainder after. Results suggest experts are better than novices across all conditions. Further, a comparison of results between the two dual task conditions indicates that attentional resources make a contribution to target tracking in both novices and experts, but only appear to make a contribution to target acquisition in novices. The results have implications for theories of multiple target tracking; the selection of radar operators; and the monitoring of their training. Copyright © 2004 John Wiley & Sons, Ltd.

P3 varies with stimulus categorization rather than probability

We examined the relationship between P3 amplitude elicited by different auditory task stimuli and response-defined categorization requirements. Two experiments were conducted which examined the effects of the categorization of tone pitches (experiment 1) and spoken digits (experiment 2) on P3 amplitude. The categorization requirements were the same in both experiments. Two patterns of results were found in experiment 1: first, when subjects had to categorize an end term within the stimulus scale the largest P3 middle tone pitch was required, P3 amplitudes were smallest for the middle tone pitch and increased in amplitude at both ends of the stimulus amplitudes were elicited by this stimulus. Secondly, when all tone pitches had to be categorized separately or when the categorization of the scale. Different patterns of results were found in experiment 2. When subjects had to categorize 1 out of 5 spoken digits the largest P3 amplitudes were elicited by this digit no matter which position it had on the stimulus scale. In addition, only negligible P3s were found in the condition in which each of the digits had to be categorized separately. It is argued (i) that spoken words yield a higher degree of distinctiveness than tone pitches and (ii) that P3 amplitude is related to the probability of response-defined task categories only to the extent that stimuli can easily be assigned to categories.

Identification and categorization of letters and digits.

Identification and categorization of letters and digits.

Identification and categorization of letters and digits.

Identification and categorization of letters and digits.

Presentation modality, rehearsal-prevention conditions, and auditory confusions in tests of short-term memory

Ss were presented four-letter sequences either auditorily or visually and asked for ordered recall after 0, 2.1, 4.2, 8.4, or 12.6 sec of digit categorization. Three different rehearsal-prevention conditions were required during presentation of the memory set: categorizing, suppressing (saying "dah"), or pronouncing each letter. Recall was worst after categorizing, best after pronouncing. Auditory presentation led to better recall after no delay but more rapid forgetting than visual presentation, regardless of the rehearsal-prevention condition. These results, and analyses of auditory confusions, are inconsistent with a view of memory which asserts that sensory information is encoded auditorily regardless of presentation modality or vocalization behavior during presentation.