Search PubMed⌕ Search

SEARCH · Search PubMed

Results for “representation learning”

Search indexed PubMed citations on genomics, clinical trials, systematic reviews and public health. Explore titles, authors and supplied subject terms, then open the PubMed record.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 649 records · Page 36Linked to original sources
PubMed2025

MO-GCAN: multi-omics integration based on graph convolutional and attention networks.

MOTIVATION: Cancer subtypes play a critical role in disease progression, prognosis, and treatment, making their detection essential for tailoring precision medicine. Studies have shown that multi-omics integration outperforms single-omics approaches in cancer subtyping tasks. However, due to the high-dimensionality of multi-omics data, many existing studies either fail to capture the correlation between true labels and learned features, or lack sufficient capacity to model complex biological representations. These limitations hinder the full potential of leveraging the rich and complementary information embedded in multi-omics datasets. RESULT: We propose a framework that leverages supervised feature learning and classification based on a graph-based learning approach with attention mechanism for cancer subtyping. More specifically, we train graph convolutional network models on each omics dataset to extract latent representations, which are then concatenated to form a comprehensive multi-omics feature embedding. We further develop sample fusion network based on the omics-specific graphs, incorporating the derived features and feeding them into a graph attention model for subtype classification. This two-stage multi-omics framework is applied to eight cancer types, with performance evaluated in terms of test accuracy, training time, macro-averaged precision, recall, and F-score. Experimental results show that the proposed method outperforms state-of-the-art approaches across various cancer types. Additionally, we provide empirical evidence supporting the hypothesis that retaining a limited number of high-confidence edges and utilizing enriched embeddings from intermediate graph neural network layers can improve predictive performance. AVAILABILITY AND IMPLEMENTATION: Data and the code are available at https://github.com/YD-00/MO-GCAN-Updated.git.

Neoplasms↗
PubMed2002

Acquisition of visual shape primitives.

Shape primitives have long been proposed as components for object models in the visual system, and account for a considerable body of behavioral findings. While a large amount of effort has been devoted to the study of detection of these parts in the scenes, no research has been undertaken simulating the acquisition of these representations. We present a model which suggests how the shape primitives may be learned by experience in a self-organized fashion. This model offers the first successful unsupervised learning of shape primitives which are as complex as object parts and can serve as intermediate representations for various objects. The algorithm uses synthetic gray-level objects, each composed of several parts (primitives or else), and shape primitives emerge as a result of partial matches between several objects. Our algorithm does not use any a priori knowledge about any attributes of the patterns to be learned; and the recurrence of these visual patterns in various objects is the only basis for their emergence as new features.

Algorithms↗
PubMed1999

Acquisition of output irregular orthographic representations in normal adults: an experimental study.

Twenty normal controls without previous knowledge of French were asked to learn 12 French words that could not be written by Italian correspondence rules. After acquisition of the phonological representations, participants were presented the pictures and asked to write the corresponding words (baseline). They were then presented 3 times with the pictures and the corresponding written words. After a filled delay of 10 min, they were re-presented the pictures and asked to write the corresponding words (testing). A week later, participants were again requested to write the words (follow-up). Number of words correctly written at testing and at follow-up significantly differs from baseline, as well as number of words correctly written at testing and at follow-up. These results are discussed and it is suggested that acquisition of irregular output orthographic representations can be supported by knowledge of orthographic representations for reading and that rehabilitation of patients with damage to output orthographic representations can utilize input orthographic representations.

Adult↗
PubMed1999

Amygdala circuitry in attentional and representational processes.

The amygdala has long been implicated in the display of emotional behavior and emotional information processing, especially in the context of aversive events. In this review, we discuss recent evidence that links the amygdala to several aspects of food-motivated associative learning, including functions often characterized as attention, reinforcement and representation. Each of these functions depends on the operation of separate amygdalar subsystems, through their connections with other brain systems. Notably, very different processing systems seem to be mediated by the central nucleus and basolateral amygdala, subregions of the amygdala that differ in their anatomy and in their connectivity. The basolateral amygdala is involved in the acquisition and representation of reinforcement value, apparently through its connections with ventral striatal dopamine systems and with the orbitofrontal cortex. The dentral nucleus, however, contributes heavily to attentional function in conditioning, by way of its influence on basal forebrain cholinergic systems and on the dorsolateral striatum.

Journal Article↗
PubMed2001

Lesions of the basolateral amygdala disrupt selective aspects of reinforcer representation in rats.

The amygdala is known to play a role in learning about motivationally significant events. We investigated this role further by examining the effects of excitotoxic lesions of the basolateral amygdala on the ability of rats to use instrumental outcomes to direct responding (the differential outcomes effect) and on the ability of Pavlovian cues to modulate instrumental performance based on shared outcomes (reinforcer-selective Pavlovian-to-instrumental transfer). We found that basolateral amygdala (BLA) lesions did not affect the ability of rats to learn a basic instrumental conditional discrimination, but did disrupt the ability of differential outcomes to facilitate acquisition. In Pavlovian-to-instrumental transfer, BLA lesions did not disrupt the basic enhancement of instrumental performance but did abolish the reinforcer specificity of that enhancement. These results suggest that the BLA is involved in the representation of the sensory aspects of motivationally significant events.

Amygdala↗
PubMed2000

Minimizing binding errors using learned conjunctive features.

We have studied some of the design trade-offs governing visual representations based on spatially invariant conjunctive feature detectors, with an emphasis on the susceptibility of such systems to false-positive recognition errors-Malsburg's classical binding problem. We begin by deriving an analytical model that makes explicit how recognition performance is affected by the number of objects that must be distinguished, the number of features included in the representation, the complexity of individual objects, and the clutter load, that is, the amount of visual material in the field of view in which multiple objects must be simultaneously recognized, independent of pose, and without explicit segmentation. Using the domain of text to model object recognition in cluttered scenes, we show that with corrections for the nonuniform probability and nonindependence of text features, the analytical model achieves good fits to measured recognition rates in simulations involving a wide range of clutter loads, word size, and feature counts. We then introduce a greedy algorithm for feature learning, derived from the analytical model, which grows a representation by choosing those conjunctive features that are most likely to distinguish objects from the cluttered backgrounds in which they are embedded. We show that the representations produced by this algorithm are compact, decorrelated, and heavily weighted toward features of low conjunctive order. Our results provide a more quantitative basis for understanding when spatially invariant conjunctive features can support unambiguous perception in multiobject scenes, and lead to several insights regarding the properties of visual representations optimized for specific recognition tasks.

Algorithms↗
PubMed2005

Relearning sound localization with a new ear.

Human sound localization results primarily from the processing of binaural differences in sound level and arrival time for locations in the horizontal plane (azimuth) and of spectral shape cues generated by the head and pinnae for positions in the vertical plane (elevation). The latter mechanism incorporates two processing stages: a spectral-to-spatial mapping stage and a binaural weighting stage that determines the contribution of each ear to perceived elevation as function of sound azimuth. We demonstrated recently that binaural pinna molds virtually abolish the ability to localize sound-source elevation, but, after several weeks, subjects regained normal localization performance. It is not clear which processing stage underlies this remarkable plasticity, because the auditory system could have learned the new spectral cues separately for each ear (spatial-mapping adaptation) or for one ear only, while extending its contribution into the contralateral hemifield (binaural-weighting adaptation). To dissociate these possibilities, we applied a long-term monaural spectral perturbation in 13 subjects. Our results show that, in eight experiments, listeners learned to localize accurately with new spectral cues that differed substantially from those provided by their own ears. Interestingly, five subjects, whose spectral cues were not sufficiently perturbed, never yielded stable localization performance. Our findings indicate that the analysis of spectral cues may involve a correlation process between the sensory input and a stored spectral representation of the subject's ears and that learning acts predominantly at a spectral-to-spatial mapping level rather than at the level of binaural weighting.

Acoustic Stimulation↗
PubMedDate not supplied

Alterations in the thickness of motor cortical subregions after motor-skill learning and exercise.

Behavioral manipulations such as housing in an enriched environment have been shown to increase brain weight and visual cortical thickness. The present study was designed to test whether skill learning or repetitive movements can alter the thickness of the motor cortex. One group of 6-mo-old Long-Evans female rats learned motor skills on an obstacle course that increased in difficulty over training and required balance and coordination. A second group ran voluntarily in exercise wheels attached to their home cage but had little opportunity for skill learning. The third group was handled daily but received no opportunity for learning or exercise. Each condition lasted 26-29 d. The skill-learning and exercise conditions had greater heart weight, and the exercise condition had greater adrenal gland weights than controls. The thickness of the motor cortex was measured in four coronal planes between -2.33 mm to -0.3 mm from bregma. Regions of interest that corresponded to published maps of forelimb and hind-limb representations were analyzed together. Rats in the skill-learning condition had significantly thicker medial cortical areas in the two anterior planes (-0.8 and -0.3 mm from bregma). These regions correspond to previously mapped hind-limb representations. The exercise group had greater thickness of the medial region at -0.8 mm from bregma. Cortical thickness in all conditions varied significantly along the medial to lateral axis. For both treatments, the effects were restricted to medial and anterior regions of interest rather than posterior or lateral regions of interest. The results indicate that robust exercise, in addition to skill learning, is capable of altering the thickness of the motor cortex, but that the effects are restricted rather than distributed within the regions studied.

Animals↗
PubMed1991

Stimulus representation: a subprocess of imprinting and conditioning.

We suggest a way to reconcile imprinting and associative learning that respects the real differences between the two phenomena but helps to recognize underlying commonalities. Rather than treating each type of learning as the manifestation of a unitary mechanism, we approach learning as a combination of separate subprocesses. Exploration of the literature regarding one of these subprocesses, namely, that governing the representation of stimuli, revealed striking similarities between imprinting and conditioning. These similarities suggest predictions for fresh experimental work that will help to uncover the general rules by which combinations of stimulus features are represented in memory.

Animals↗
PubMed2000

Disparity tuning as simulated by a neural net.

Previous research has suggested that the processing of binocular disparity in complex cells may be described with an energy formalism. The energy formalism allows for a representation of disparity by differences in the position or in the phase of monocular receptive subfields of binocular cells, or by combination of these two types. We studied the coding of disparities with an approach complementary to previous algorithmic investigations. Since realization of these representations is probably not genetically determined but learned during ontogeny, we used backpropagation networks to study which of these three possibilities were realized within neural nets. Three types of networks were trained with noise patterns in analogy to the three types of energy models. The networks learned the task and generalized to untrained correlated noise pattern input. Outputs were broadly tuned to spatial frequency and did not respond to anti-correlated noise patterns. Although the energy model was not explicitly implemented, we could analyze the outputs of the networks using predictions of the energy formalism. After learning was completed, the model neurons preferred position shifts over phase shifts in representing disparity. We discuss the general meaning of these findings and the correspondences and deviations between the energy model, V1 neurons, and our networks.

Computer Simulation↗
PubMed2006

Intentions and expectations in temporal binding.

Recently, it has been shown that the perceived times of voluntary movements and their effects are perceived as shifted towards each other. This temporal binding phenomenon was explained by an integrated representation of movement and effect, facilitating operant learning and the experience of intentionality. Here, we investigated whether temporal binding depends on explicit intentional attributions. In Experiment 1, participants intended to either produce or avoid producing an effect (a tone) by the timing of their movements, with the ratio of success being fixed at 2:1. In Experiments 2 and 3, the influence of the action-effect contingency ratio on temporal binding was controlled for by removing the intentional attribution of the effect. The results indicate that temporal binding is a general associative mechanism that facilitates the learning of movement-effect contingencies. Beyond that, temporal binding is sensitive to explicit intentional attributions, which selectively enhance the link between an intentional movement and the effect a moving agent intends to produce.

Analysis of Variance↗
PubMed2004

Parallel processing across neural systems: implications for a multiple memory system hypothesis.

A common conceptualization of the organization of memory systems in brain is that different types of memory are mediated by distinct neural systems. Strong support for this view comes from studies that show double (or triple) dissociations between spatial, response, and emotional memories following selective lesions of hippocampus, striatum, and the amygdala. Here, we examine the extent to which hippocampal and striatal neural activity patterns support the multiple memory systems view. A comparison is made between hippocampal and striatal neural correlates with behavior during asymptotic performance of spatial and response maze tasks. Location- (or place), movement, and reward-specific firing patterns were found in both structures regardless of the task demands. Many, but not all, place fields of hippocampal and striatal neurons were similarly affected by changes in the visual and reward context regardless of the cognitive demands. Also, many, but not all, hippocampal and striatal movement-sensitive neurons showed significant changes in their behavioral correlates after a change in visual context, irrespective of cognitive strategy. Similar partial reorganization was observed following manipulations of the reward condition for cells recorded from both structures, again regardless of task. Assuming that representations that persist across context changes reflect learned information, we make the following conclusions. First, the consistent pattern of partial reorganization supports a view that the analysis of spatial, response, and reinforcement information is accomplished via an error-driven, or match-mismatch, algorithm across neural systems. Second, task-relevant processing occurs continuously within hippocampus and striatum regardless of the cognitive demands of the task. Third, given the high degree of parallel processing across allegedly different memory systems, we propose that different neural systems may effectively compete for control of a behavioral expression system. The strength of the influence of any one neural system on behavioral output is likely modulated by factors such as motivation, experience, or hormone status.

Animals↗
PubMed2005

Neural systems recruited by drug- and food-related cues: studies of gene activation in corticolimbic regions.

In order to survive, animals must acquire information about the reward value of stimuli in their environment. This process partly depends on the ability of the organism to make associations between the environmental context and the internal representation of value. While this type of learning probably evolved in order to promote behaviors that increase fitness (e.g., ingestive and sexual behavior), neuropsychological research utilizing addictive drugs, which are potent artificial reinforcers, has led to a deeper understanding of reinforcement mechanisms. Through these associations, sensory cues can acquire emotional salience and motivational properties. Exposure to drug-related cues in human addicts results in drug craving and localized activation of central circuits that are known to mediate cue-induced reinstatement of drug-seeking behavior in animal models of relapse. Similar regional activation patterns occur in humans in response to cues associated with foods. Furthermore, drug- and food-related cues not only activate common neuroanatomical regions but also result in similar activity-regulated gene expression programs within these shared areas. Here we discuss recent studies from our laboratory that investigate gene expression patterns elicited by exposure to palatable food- or drug-related cues. These studies suggest that the central nervous system stores and utilizes information about 'natural' and drug reinforcers in similar ways, both neuroanatomically and biochemically. These considerations may have important implications for the pharmacological and cognitive-behavioral treatments of substance use disorders, addiction, eating disorders, and obesity.

Animals↗
PubMed2005

Allocentric and egocentric spatial impairments in a case of topographical disorientation.

We describe a patient with a topographical disorientation after a stroke of the right mediotemporooccipital lobe including the parahippocampal cortex (PHC). Clinical observations and neuropsychological testing reveal an impairment of allocentric spatial representations as well as impairments of visuospatial learning and memory. These findings are in accordance with the well-known function of the PHC in topographical disorientation. As a new finding, results from oculomotor tasks show additional impairments of the egocentric spatial coordinate frame suggesting that in topographical disorientation due to a lesion of the right mediotemporooccipital lobe not only allocentric but also egocentric visuospatial functions are disturbed.

Aged↗
PubMed2000

Adaptive information processing in microtubule networks.

Microtubule networks provide a wide range of microskeletal and micromuscular functionalities. Evidence from a number of directions suggests that they can also serve as a medium for intracellular signaling processing. The model presented here comprises an empirically motivated representation of microtubule growth dynamics, an abstract representation of signal processing, and a feedback learning mechanism that we refer to as adaptive self-stabilization. The growth model mimics the dynamic instability picture of microtubule formation and decomposition, but as modulated by the binding activity of microtubule associated proteins (or MAPs). The signal processing submodel treats each microtubule as a string of linked discrete oscillators capable of propagating signals that are introduced, manipulated, and extracted by bound MAP activity. Adaptive self-stabilization is essentially feedback acting on signal processing capabilities via the growth dynamics. The network is presented with a training set of patterns. If the input-output behavior is satisfactory MAP binding affinity increases, thereby stabilizing the network structure; otherwise the binding affinity decreases, allowing for more structural variation. The results obtained suggest that adaptive capabilities are practically inevitable in microtubule networks, a conclusion strengthened by the fact that the signal processing and growth dynamics mechanisms available in nature are undoubtedly much richer than those represented in the model.

Adaptation, Physiological↗
PubMed2001

Medical informatics: reasoning methods.

The progress of medical informatics has been characterized by the development of a wide range of reasoning methods. These reasoning methods are based on organizing principles that make use of the various relations existing in medical domains: associations, probabilities, causality, functional relationships, temporal relations, locality, similarity, and clinical practice. Some, such as those based on associations and probabilities have been developed to the point where there are off-the-shelf tools available for the researcher to develop new decision support tools. Others such as temporal relations require more effort to use effectively. Even so, we have learned the importance of a separate explicit representation of the domain knowledge and have considerable experience and an impressive armamentarium with which to face the new milieu provided by the Internet.

Artificial Intelligence↗
PubMed1991

Syntax and semantics in the acquisition of locative verbs.

Children between the ages of three and seven occasionally make errors with locative verbs like pour and fill, such as *I filled water into the glass and *I poured the glass with water (Bowerman, 1982). To account for this pattern of errors, and for how they are eventually unlearned, we propose that children use a universal linking rule called object affectedness: the direct object corresponds to the argument that is specified as 'affected' in some particular way in the semantic representation of a verb. However, children must learn which verbs specify which of their arguments as being affected; specifically, whether it is the argument whose referent is undergoing a change of location, such as the content argument of pour, or the argument whose referent is undergoing a change of state, such as the container argument of fill. This predicts that syntactic errors should be associated with specific kinds of misinterpretations of verb meaning. Two experiments were performed on the ability of children and adults to understand and produce locative verbs. The results confirm that children tend to make syntactic errors with sentences containing fill and empty, encoding the content argument as direct object (e.g. fill the water). As predicted, children also misinterpreted the meanings of fill and empty as requiring not only that the container be brought into a full or empty state, but also that the content move in some specific manner (by pouring, or by dumping). Furthermore, children who misinterpreted the verbs' meanings were more likely to make syntactic errors with them. These findings support the hypothesis that verb meaning and syntax are linked in precise ways in the lexicons of language learners.

Child↗
PubMed2006

Hierarchical encoding of behavior: translating perception into action.

People encode goal-directed behaviors, such as assembling an object, by segmenting them into discrete actions, organized as goal-subgoal hierarchies. Does hierarchical encoding contribute to observational learning? Participants in 3 experiments segmented an object assembly task into coarse and fine units of action and later performed it themselves. Hierarchical encoding, measured by segmentation patterns, correlated with more accurate and more hierarchically structured performance of the later assembly task. Furthermore, hierarchical encoding increased when participants (a) segmented coarse units first, (b) explicitly looked for hierarchical structure, and (c) described actions while segmenting them. Improving hierarchical encoding always led to improvements in learning, as well as a surprising shift toward encoding and executing actions from the actor's spatial perspective instead of the participants' own. Hierarchical encoding facilitates observational learning by organizing perceived actions into a representation that can serve as an action plan.

Awareness↗