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 163 records · Page 9Linked to original sources
PubMed2004

Configural representations in transverse patterning with a hippocampal model.

The hippocampus is necessary in both humans and rats for learning configural representations in tasks such as transverse patterning. The transverse patterning task, (A+B-, B+C-, C+A-), requires representing individual stimuli in the context of other stimuli. This paper extends a previous application to rat data [INNS World Congress on Neural Networks, 1995; Biol Cybern 6 (1998a) 203] by applying a model of the CA3 region of the hippocampus to human data. A decision function is also added that enables the system to choose among training items. Analysis of the simulations show that configural representations are formed by unique neural codes that depend on temporal and stimuli context. Based on the simulations, we hypothesize that configural representations in biological networks depend on a proper balance of input and context representations. Furthermore, the division of labor between functions in the model is a specific working hypothesis that in learning this task the hippocampus specializes in sequence prediction and the decision function evaluates the predictions.

Hippocampus↗
PubMed2006

fMRI investigation of cortical and subcortical networks in the learning of abstract and effector-specific representations of motor sequences.

A visuo-motor sequence can be learned as a series of visuo-spatial cues or as a sequence of effector movements. Earlier imaging studies have revealed that a network of brain areas is activated in the course of motor sequence learning. However, these studies do not address the question of the type of representation being established at various stages of visuo-motor sequence learning. In an earlier behavioral study, we demonstrated that acquisition of visuo-spatial sequence representation enables rapid learning in the early stage and progressive establishment of somato-motor representation helps speedier execution by the late stage. We conducted functional magnetic resonance imaging (fMRI) experiments wherein subjects learned and practiced the same sequence alternately in normal and rotated settings. In one rotated setting (visual), subjects learned a new motor sequence in response to an identical sequence of visual cues as in normal. In another rotated setting (motor), the display sequence was altered as compared to normal, but the same sequence of effector movements was used to perform the sequence. Comparison of different rotated settings revealed analogous transitions both in the cortical and subcortical sites during visuo-motor sequence learning-a transition of activity from parietal to parietal-premotor and then to premotor cortex and a concomitant shift was observed from anterior putamen to a combined activity in both anterior and posterior putamen and finally to posterior putamen. These results suggest a putative role for engagement of different cortical and subcortical networks at various stages of learning in supporting distinct sequence representations.

Adult↗
PubMed2000

A neural network architecture for visual selection.

This article describes a parallel neural net architecture for efficient and robust visual selection in generic gray-level images. Objects are represented through flexible star-type planar arrangements of binary local features which are in turn star-type planar arrangements of oriented edges. Candidate locations are detected over a range of scales and other deformations, using a generalized Hough transform. The flexibility of the arrangements provides the required invariance. Training involves selecting a small number of stable local features from a predefined pool, which are well localized on registered examples of the object. Training therefore requires only small data sets. The parallel architecture is constructed so that the Hough transform associated with any object can be implemented without creating or modifying any connections. The different object representations are learned and stored in a central module. When one of these representations is evoked, it "primes" the appropriate layers in the network so that the corresponding Hough transform is computed. Analogies between the different layers in the network and those in the visual system are discussed. Furthermore, the model can be used to explain certain experiments on visual selection reported in the literature.

Algorithms↗
PubMed2004

Active navigation and orientation-free spatial representations.

In this study, we examined the orientation dependency of spatial representations following various learning conditions. We assessed the spatial representations of human participants after they had learned a complex spatial layout via map learning, via navigating within a real environment, or via navigating through a virtual simulation of that environment. Performances were compared between conditions involving (1) multiple- versus single-body orientation, (2) active versus passive learning, and (3) high versus low levels of proprioceptive information. Following learning, the participants were required to produce directional judgments to target landmarks. Results showed that the participants developed orientation-specific spatial representations following map learning and passive learning, as indicated by better performance when tested from the initial learning orientation. These results suggest that neither the number of vantage points nor the level of proprioceptive information experienced are determining factors; rather, it is the active aspect of direct navigation that leads to the development of orientation-free representations.

Adolescent↗
PubMed1994

Implicit learning.

Implicit learning is nonepisodic learning of complex information in an incidental manner, without awareness of what has been learned. Implicit learning experiments use 3 different stimulus structures (visual, sequence, and function) and 3 different dependent measures or response modalities (conceptual fluency, efficiency, and prediction and control). Implicit learning may require a certain minimal amount of attention and may depend on attentional and working memory mechanisms. The result of implicit learning is implicit knowledge in the form of abstract (but possibly instantiated) representations rather than verbatim or aggregate representations. Implicit learning shows biases and dissociations in learning different stimulus structures. The dependence of implicit learning on particular brain areas is discussed, some conclusions are drawn for modeling implicit learning, and the interaction of implicit and explicit learning is considered.

Concept Formation↗
PubMed1994

Precursors of linguistic knowledge.

It is agreed that infants require pre-existing conceptual meanings to learn language, although little is known about what those meanings are. By default they have been assumed to be the sensorimotor schemas described by Piaget. However, sensorimotor schemas are not concepts and are not the right sort of representation for learning language. Recent research shows that along with sensorimotor schemas infants are simultaneously developing a rich conceptual system. I have proposed that these concepts are represented by sets of image-schemas, each of which represents a meaning. Image-schemas are created by a process of perceptual analysis that redescribes perceptual information into simplified spatial representations. These representations allow language to be learned by providing an analogue-digital interface between the continuous process of perception and the discrete propositional forms of language.

Child↗
PubMed2005

Specificity of sensorimotor learning and the neural code: neuronal representations in the primary motor cortex.

Human studies show that the learning of a new sensorimotor mapping that requires adaptation to directional errors is local and generalizes poorly to untrained directions. We trained monkeys to learn new visuomotor rotations for only one target in space and recorded neuronal activity in the primary motor cortex before, during and after learning. Similar to humans, the monkeys showed poor transfer of learning to other directions, as observed by behavioral aftereffects for untrained directions. To test for internal representations underlying these changes, we compared two features of neuronal activity before and after learning: changes in firing rates and changes in information content. Specific elevations of firing rate were only observed in a subpopulation of cells in the motor cortex with directional properties corresponding to the locally learned rotation; namely cells only showed plasticity if their preferred direction was near the training one. We applied measures from information theory to probe for learning-related changes in the neuronal code. Single cells conveyed more information about the direction of movement and this specific improvement in encoding was correlated with an increase in the slope of the neurons' tuning curve. Further, the improved information after learning enabled a more accurate reconstruction of movement direction from neuronal populations. Our findings suggest a neural mechanism for the confined generalization of a newly acquired internal model by showing a tight relationship between the locality of learning and the properties of neurons. They also provide direct evidence for improvement in the neural code as a result of learning.

Action Potentials↗
PubMed1997

Sensory modality specificity of neural activity related to memory in visual cortex.

Previous studies have shown that when monkeys perform a delayed match-to-sample (DMS) task, some neurons in inferotemporal visual cortex are activated selectively during the delay period when the animal must remember particular visual stimuli. This selective delay activity may be involved in short-term memory. It does not depend on visual stimulation: both auditory and tactile stimuli can trigger selective delay activity in inferotemporal cortex when animals expect to respond to visual stimuli in a DMS task. We have examined the overall modality specificity of delay period activity using a variety of auditory/visual cross-modal and unimodal DMS tasks. The cross-modal DMS tasks involved making specific long-term memory associations between visual and auditory stimuli, whereas the unimodal DMS tasks were standard identity matching tasks. Delay activity existed in auditory/visual cross-modal DMS tasks whether the animal anticipated responding to visual or auditory stimuli. No evidence of selective delay period activation was seen in a purely auditory DMS task. Delay-selective cells were relatively common in one animal where they constituted up to 53% neurons tested with a given task. This was only the case for up to 9% of cells in a second animal. In the first animal, a specific long-term memory representation for learned cross-modal associations was observed in delay activity, indicating that this type of representation need not be purely visual. Furthermore, in this same animal, delay activity in one cross-modal task, an auditory-to-visual task, predicted correct and incorrect responses. These results suggest that neurons in inferotemporal cortex contribute to abstract memory representations that can be activated by input from other sensory modalities, but these representations are specific to visual behaviors.

Acoustic Stimulation↗
PubMed1995

Joint coordination deficits in limb apraxia.

Competing models of the basis of limb apraxia were tested through analysis of joint coordination deficits in three apraxic subjects with lesions that included the left parietal lobe. Three-dimensional shoulder, elbow, wrist and hand trajectories were recorded for repetitive 'slicing' gestures made in a series of conditions in which contextual cues were introduced in a graded fashion. The apraxic subjects showed marked deficits in joint coordination across context conditions. Even when actually manipulating a tool and object, the apraxic subjects failed to show proper joint synchronization, failed to apportion their relative joint amplitudes properly, and failed to produce the correct phase relationships among pairs of arm angles. Thus, apraxic subjects not only have deficits in the spatial plan for the movement, but they also have deficits in translating those plans into the details of the angular motions at the joints, even when actually manipulating a tool and object. These data support a model of apraxia in which apraxia can result from either the destruction of visuo-kinaesthetic motor representations of learned movement, stored in posterior association cortex, or from a separation of these representations from premotor or motor areas.

Aged↗
PubMed1999

Implicit learning in 3D object recognition: the importance of temporal context.

A novel architecture and set of learning rules for cortical self-organization is proposed. The model is based on the idea that multiple information channels can modulate one another's plasticity. Features learned from bottom-up information sources can thus be influenced by those learned from contextual pathways, and vice versa. A maximum likelihood cost function allows this scheme to be implemented in a biologically feasible, hierarchical neural circuit. In simulations of the model, we first demonstrate the utility of temporal context in modulating plasticity. The model learns a representation that categorizes people's faces according to identity, independent of viewpoint, by taking advantage of the temporal continuity in image sequences. In a second set of simulations, we add plasticity to the contextual stream and explore variations in the architecture. In this case, the model learns a two-tiered representation, starting with a coarse view-based clustering and proceeding to a finer clustering of more specific stimulus features. This model provides a tenable account of how people may perform 3D object recognition in a hierarchical, bottom-up fashion.

Animals↗
PubMed2005

Learning-induced plasticity of cortical representations does not affect GAD65 mRNA expression and immunolabeling of cortical neuropil.

Two forms of glutamic acid decarboxylase (GAD) are present in inhibitory neurons of the mammalian brain, a 65-kDa isoform (GAD65) and a 67-kDa isoform (GAD67). We have previously found that GAD67 is upregulated during learning-dependent plasticity of cortical vibrissal representations of adult mice. After sensory conditioning involving pairing stimulation of vibrissae with a tail shock, the increase in mRNA expression and density of GAD67-immunoreactive neurons was observed in barrels representing vibrissae activated during the training. In the present study, using the same experimental model, we examined GAD65 mRNA and protein levels in the barrel cortex. For this purpose, we used in situ hybridization and immunohistochemistry. No changes in the level of GAD65 mRNA expression were detected after the training. The pattern of GAD65 mRNA expression was complementary to that observed for GAD67. Immunocytochemical analysis found no changes in immunolabeling of neuropil of the barrels representing the vibrissae activated during the training. The results show that, in contrast to GAD67, cortical plasticity induced by sensory learning does not affect the expression of GAD65.

Animals↗
PubMed1993

Cortical plasticity and memory.

The roles of extrinsically modulated, plastic Hebb-like synapses and dynamic cortical cell assemblies underlying cortical plasticity in learning and memory operations are described. From our understanding of the distributed form of representation of learned behaviors in somatosensory and auditory cortical fields, and given new findings about the nature and distribution of responses representing learned and remembered stimuli in the inferior temporal cortex, a hypothetical picture of the cortical engram representing learned behaviors and memories is posited.

Afferent Pathways↗
PubMed2003

A neural model of how the brain represents and compares multi-digit numbers: spatial and categorical processes.

Both animals and humans represent and compare numerical quantities, but only humans have evolved multi-digit place-value number systems. This article develops a Spatial Number Network, or SpaN, model to explain how these shared numerical capabilities are computed using a spatial representation of number quantities in the Where cortical processing stream, notably the inferior parietal cortex. Multi-digit numerical representations that obey a place-value principle are proposed to arise through learned interactions between categorical language representations in the What cortical processing stream and the Where spatial representation. Learned semantic categories that symbolize separate digits, as well as place markers like 'ty,' 'hundred,' and 'thousand,' are associated through learning with the corresponding spatial locations of the Where representation. Such What-to-Where auditory-to-visual learning generates place-value numbers as an emergent property, and may be compared with other examples of multi-modal cross-modality learning, including synesthesia. The model quantitatively simulates error rates in quantification and numerical comparison tasks, and reaction times for number priming and numerical assessment and comparison tasks. In the Where cortical process, transient responses to inputs are integrated before they activate an ordered spatial map that selectively responds to the number of events in a sequence and exhibits Weber law properties. Numerical comparison arises from activity pattern changes across the spatial map that define a 'directional comparison wave.' Variants of these model mechanisms have elsewhere been used to explain data about other Where stream phenomena, such as motion perception, spatial attention, and target tracking. The model is compared with other models of numerical representation.

Animals↗
PubMed2001

Chunking mechanisms in human learning.

Pioneering work in the 1940s and 1950s suggested that the concept of 'chunking' might be important in many processes of perception, learning and cognition in humans and animals. We summarize here the major sources of evidence for chunking mechanisms, and consider how such mechanisms have been implemented in computational models of the learning process. We distinguish two forms of chunking: the first deliberate, under strategic control, and goal-oriented; the second automatic, continuous, and linked to perceptual processes. Recent work with discrimination-network computational models of long- and short-term memory (EPAM/CHREST) has produced a diverse range of applications of perceptual chunking. We focus on recent successes in verbal learning, expert memory, language acquisition and learning multiple representations, to illustrate the implementation and use of chunking mechanisms within contemporary models of human learning.

Journal Article↗
PubMed1982

Pretraining septal driving of hippocampal rhythmic slow activity facilitates acquisition of visual discrimination.

Studies have shown hippocampal rhythmic slow activity (H-RSA) to reflect both learning and memory processes across a variety of species and conditioning procedures. In order to investigate these relations further. H-RSA was manipulated by medial septal (MS) stimulation directly before training rats for light/dark discrimination in a T-maze. Rats that had H-RSA increased learned the discrimination significantly faster than those that had H-RSA blocked and control rats. In addition, increase in H-RSA before training was found to correlate with speed of learning. The evidence of this study is consistent with results of other studies showing septal-hippocampal interaction during learning. The results also support the view that H-RSA may be a neurophysiological representation of learning and memory processes.

Animals↗
PubMed1996

A primate genesis model of focal dystonia and repetitive strain injury: I. Learning-induced dedifferentiation of the representation of the hand in the primary somatosensory cortex in adult monkeys.

In this study we tested a neuroplasticity/learning origins hypothesis for repetitive strain injuries (RSIs), including occupationally induced focal dystonia. Repetitive movements produced in a specific form and in an appropriate behavioral context cause a degradation of the sensory feedback information controlling fine motor movements, resulting in the "learned" genesis of RSIs. Two adult New World owl monkeys were trained at a behavioral task that required them to maintain an attended grasp on a hand grip that repetitively and rapidly (20 msec) opened and closed over short distances. The monkeys completed 300 behavioral trials per day (1,100 to 3,000 movement events) with an accuracy of 80 to 90%. A movement control disorder was recorded in both monkeys. Training was continued until the performance accuracy dropped to below 50%. We subsequently conducted an electrophysiologic mapping study of the representations of the hand within the primary somatosensory (SI) cortical zone. The hand representation in the true primary somatosensory cortical field, SI area 3b, was found to be markedly degraded in these monkeys, as characterized by (1) a dedifferentiation of cortical representations of the skin of the hand manifested by receptive fields that were 10 to 20 times larger than normal, (2) the emergence of many receptive fields that covered the entire glabrous surface of individual digits or that extended across the surfaces of two or more digits, (3) a breakdown of the normally sharply segregated area 3b representations of volar glabrous and dorsal hairy skin of the hand, and (4) a breakdown of the local shifted-overlap receptive field topography of area 3b, with many digital receptive fields overlapping the fields of neurons sampled in cortical penetrations up to more than four times farther apart than normal. Thus, rapid, repetitive, highly stereotypic movements applied in a learning context can actively degrade cortical representations of sensory information guiding fine motor hand movements. This cortical plasticity/learning-based dedifferentiation of sensory feedback information from the hand contributes to the genesis of occupationally derived repetitive strain injuries, including focal dystonia of the hand. Successful treatment of patients with RSI will plausibly require learning-based restoration of differentiated representations of sensory feedback information from the hand.

Animals↗
PubMed2002

The prefrontal cortex: categories, concepts and cognition.

The ability to generalize behaviour-guiding principles and concepts from experience is key to intelligent, goal-directed behaviour. It allows us to deal efficiently with a complex world and to adapt readily to novel situations. We review evidence that the prefrontal cortex-the cortical area that reaches its greatest elaboration in primates-plays a central part in acquiring and representing this information. The prefrontal cortex receives highly processed information from all major forebrain systems, and neurophysiological studies suggest that it synthesizes this into representations of learned task contingencies, concepts and task rules. In short, the prefrontal cortex seems to underlie our internal representations of the 'rules of the game'. This may provide the necessary foundation for the complex behaviour of primates, in whom this structure is most elaborate.

Animals↗
PubMed1997

Understanding time series networks: a case study in rule extraction.

A significant limitation of neural networks is that the representations they learn are usually incomprehensible to humans. We have developed an algorithm, called TREPAN, for extracting comprehensible, symbolic representations from trained neural networks. Given a trained network, TREPAN produces a decision tree that approximates the concept represented by the network. In this article, we discuss the application of TREPAN to a neural network trained on a noisy time series task: predicting the Dollar-Mark exchange rate. We present experiments that show that TREPAN is able to extract a decision tree from this network that equals the network in terms of predictive accuracy, yet provides a comprehensible concept representation. Moreover, our experiments indicate that decision trees induced directly from the training data using conventional algorithms do not match the accuracy nor the comprehensibility of the tree extracted by TREPAN.

Artificial Intelligence↗