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 253 records · Page 14Linked to original sources
PubMed2005

Cortico-hippocampal interaction and adaptive stimulus representation: a neurocomputational theory of associative learning and memory.

Computational models of the hippocampal region link psychological theories of associative learning with their underlying physiological and anatomical substrates. Our approach to theory development began with a broad description of the computations that depend on the hippocampal region in classical conditioning (Gluck and Myers, 1993 and Gluck and Myers, 2001). In this initial model, the hippocampal region was treated as an Information-processing system that transformed stimulus representations, compressing (making more similar) representations of inputs that co-occur or are otherwise redundant, while differentiating (or making less similar) representations of inputs that predict different future events. This model led to novel predictions for the behavioral consequences of hippocampal-region lesions in rodents and of brain damage in humans who have amnesia or are in the earliest stages of Alzheimer's disease. Many of these predictions have, since been confirmed by our lab and others. Functional brain imaging studies have provided further supporting evidence. In more recent computational modeling, we have shown how some aspects of this proposed information-processing function could emerge from known anatomical and physiological characteristics of the hippocampal region, including the entorhinal cortex and the septo-hippocampal cholinergic system. The modeling to date lays the groundwork for future directions that increase the depth of detail of the biological modeling, as well as the breadth of behavioral phenomena addressed. In particular, we are working now to reconcile these kinds of incremental associative learning models with other models of the hippocampal region that account for the rapid formation of declarative memories.

Animals↗
PubMed1997

Analogical transfer is effective in a serial reaction time task in Parkinson's disease: evidence for a dissociable form of sequence learning.

Several studies of procedural learning in Parkinson's disease (PD) have demonstrated that these patients are impaired with respect to age-matched control subjects. In order to examine more closely the specific impairment, we considered three dimensions along which a procedural learning task could vary. These are: (1) implicit vs explicit learning, (2) instance vs rule learning, and (3) learning with internal vs external error correction. We consider two hypotheses that could explain the impairments observed in PD for different types of explicit motor learning: (H1) an impairment related to the acquisition of rules vs specific instances, and (H2) an impairment in learning when no explicit error feedback is provided. In order to examine the condition of rule learning with external error feedback, we developed a modified version of the serial reaction time (SRT) protocol that tests analogical transfer in sequence learning (ATSL). Reaction times are measured for responses to visual stimuli that appear in several different repeating sequences. While these isomorphic sequences are different, they share a common rule. Verbatim learning of a sequence would result in negative transfer from one sequence to a different one, while rule learning would result in positive transfer. Parkinson's patients and age-matched controls demonstrate significant acquisition and positive transfer of the rule between sequences. Our results demonstrate that PD patients are capable of learning and transferring rule or schema-based representations in an explicit learning format, and that this form of learning may be functionally distinct from learning mechanisms that rely on representations of the verbatim or statistical structure of sequences.

Adult↗
PubMed1998

The acquisition of skilled motor performance: fast and slow experience-driven changes in primary motor cortex.

Behavioral and neurophysiological studies suggest that skill learning can be mediated by discrete, experience-driven changes within specific neural representations subserving the performance of the trained task. We have shown that a few minutes of daily practice on a sequential finger opposition task induced large, incremental performance gains over a few weeks of training. These gains did not generalize to the contralateral hand nor to a matched sequence of identical component movements, suggesting that a lateralized representation of the learned sequence of movements evolved through practice. This interpretation was supported by functional MRI data showing that a more extensive representation of the trained sequence emerged in primary motor cortex after 3 weeks of training. The imaging data, however, also indicated important changes occurring in primary motor cortex during the initial scanning sessions, which we proposed may reflect the setting up of a task-specific motor processing routine. Here we provide behavioral and functional MRI data on experience-dependent changes induced by a limited amount of repetitions within the first imaging session. We show that this limited training experience can be sufficient to trigger performance gains that require time to become evident. We propose that skilled motor performance is acquired in several stages: "fast" learning, an initial, within-session improvement phase, followed by a period of consolidation of several hours duration, and then "slow" learning, consisting of delayed, incremental gains in performance emerging after continued practice. This time course may reflect basic mechanisms of neuronal plasticity in the adult brain that subserve the acquisition and retention of many different skills.

Adult↗
PubMed1991

A neural network approach to hippocampal function in classical conditioning.

Hippocampal participation in classical conditioning in terms of Grossberg's (1975) attentional theory is described. According to the present rendition of this theory, pairing of a conditioned stimulus (CS) with an unconditioned stimulus (US) causes both an association of the sensory representation of the CS with the US (conditioned reinforcement learning) and an association of the sensory representation of the CS with the drive representation of the US (incentive motivation learning). Sensory representations compete among themselves for a limited-capacity short-term memory (STM) that is reflected in a long-term memory storage. The STM regulation hypothesis, which proposes that the hippocampus controls incentive motivation, self-excitation, and competition among sensory representations thereby regulating the contents of a limited capacity STM, is introduced. Under the STM regulation hypothesis, nodes and connections in Grossberg's neural network are mapped onto regional hippocampal-cerebellar circuits. The resulting neural model provides (a) a framework for understanding the dynamics of information processing and storage in the hippocampus and cerebellum during classical conditioning of the rabbit's nictitating membrane, (b) principles for understanding the effect of different hippocampal manipulations on classical conditioning, and (c) numerous novel and testable predictions.

Animals↗
PubMed1997

Empirical investigation of the benefits of partial Lamarckianism.

Genetic algorithms (GAs) are very efficient at exploring the entire search space; however, they are relatively poor at finding the precise local optimal solution in the region in which the algorithm converges. Hybrid GAs are the combination of improvement procedures, which are good at finding local optima, and GAs. There are two basic strategies for using hybrid GAs. In the first, Lamarckian learning, the genetic representation is updated to match the solution found by the improvement procedure. In the second, Baldwinian learning, improvement procedures are used to change the fitness landscape, but the solution that is found is not encoded back into the genetic string. This paper examines the issue of using partial Lamarckianism (i.e., the updating of the genetic representation for only a percentage of the individuals), as compared to pure Lamarckian and pure Baldwinian learning in hybrid GAs. Multiple instances of five bounded nonlinear problems, the location-allocation problem, and the cell formation problem were used as test problems in an empirical investigation. Neither a pure Lamarckian nor a pure Baldwinian search strategy was found to consistently lead to quicker convergence of the GA to the best known solution for the series of test problems. Based on a minimax criterion (i.e., minimizing the worst case performance across all test problem instances), the 20% and 40% partial Lamarckianism search strategies yielded the best mixture of solution quality and computational efficiency.

Algorithms↗
PubMed2004

Simulated apoptosis/neurogenesis regulates learning and memory capabilities of adaptive neural networks.

Characterization of neuronal death and neurogenesis in the adult brain of birds, humans, and other mammals raises the possibility that neuronal turnover represents a special form of neuroplasticity associated with stress responses, cognition, and the pathophysiology and treatment of psychiatric disorders. Multilayer neural network models capable of learning alphabetic character representations via incremental synaptic connection strength changes were used to assess additional learning and memory effects incurred by simulation of coordinated apoptotic and neurogenic events in the middle layer. Using a consistent incremental learning capability across all neurons and experimental conditions, increasing the number of middle layer neurons undergoing turnover increased network learning capacity for new information, and increased forgetting of old information. Simulations also showed that specific patterns of neural turnover based on individual neuronal connection characteristics, or the temporal-spatial pattern of neurons chosen for turnover during new learning impacts new learning performance. These simulations predict that apoptotic and neurogenic events could act together to produce specific learning and memory effects beyond those provided by ongoing mechanisms of connection plasticity in neuronal populations. Regulation of rates as well as patterns of neuronal turnover may serve an important function in tuning the informatic properties of plastic networks according to novel informational demands. Analogous regulation in the hippocampus may provide for adaptive cognitive and emotional responses to novel and stressful contexts, or operate suboptimally as a basis for psychiatric disorders. The implications of these elementary simulations for future biological and neural modeling research on apoptosis and neurogenesis are discussed.

Animals↗
PubMed2006

Computational models of the hippocampal region: implications for prediction of risk for Alzheimer's disease in non-demented elderly.

We have pursued an interdisciplinary research program to develop novel behavioral assessment tools for evaluating specific memory impairments following damage to the medial temporal lobe, including the hippocampus and associated structures that show pathology early in the course of Alzheimer's disease (AD). Our approach uses computational models to identify the functional consequences of hippocampal-region damage, leading to testable predictions in both rodents and humans. Our modeling argues that hippocampal-region dysfunction may selectively impair the ability to generalize when familiar information is presented in novel recombinations. Previous research has shown that specific reductions in hippocampal volume in non-demented elderly individuals correlate with future development of AD. In two previous studies, we tested non-demented elderly with and without mild hippocampal atrophy (HA) on stimulus-response learning tasks. Individuals with and without HA could learn the initial information, but the HA group was selectively impaired on transfer tests where familiar features and objects were recombined. This suggests that such generalization deficits may be behavioral markers of HA, and an early indicator of risk for subsequent cognitive decline. Converging support for the relevance of these tasks to aging and Alzheimer's disease comes from our recent fMRI studies of individuals with mild cognitive impairment (MCI). Activity in the hippocampus declines with progressive training on these tasks, suggesting that the hippocampus is important for learning new stimulus representations that support subsequent transfer. Individuals with HA may be able to learn, but in a more hippocampal-independent fashion that does not support later transfer. Ultimately, this line of research could lead to a novel battery of behavioral tests sensitive to very mild hippocampal atrophy and risk for decline to AD, allowing early diagnosis and also allowing researchers to test new Alzheimer's drugs that target individuals in the earliest stages of the disease - before significant cognitive decline. A new mouse version of one of our tasks shows promise for translating these paradigms into rodents, allowing for future studies of therapeutic interventions in transgenic mouse models of AD.

Aged↗
PubMed2002

Normal functional imaging of the basal ganglia.

In non-human primates, all cortical projections into the striatum are topographically organized in discrete parallel circuits. These circuits are involved in distinct behavioral functions. In humans, functional imaging data support a similar organization of the striatum. The representation of the different elements of a motor act (such as movement learning, selection, mental representation, preparation and execution) are represented differentially along distinct cortico-basal ganglia circuits. Saccadic eye movements are predominantly represented in the caudate nucleus. Motivational processes are represented in the ventral part of the striatum. Thus, functional imaging data suggest that the human striatum is organized in parallel motor circuits similar to those found in animal studies.

Animals↗
PubMed2004

Learning to detect objects in images via a sparse, part-based representation.

We study the problem of detecting objects in still, gray-scale images. Our primary focus is the development of a learning-based approach to the problem that makes use of a sparse, part-based representation. A vocabulary of distinctive object parts is automatically constructed from a set of sample images of the object class of interest; images are then represented using parts from this vocabulary, together with spatial relations observed among the parts. Based on this representation, a learning algorithm is used to automatically learn to detect instances of the object class in new images. The approach can be applied to any object with distinguishable parts in a relatively fixed spatial configuration; it is evaluated here on difficult sets of real-world images containing side views of cars, and is seen to successfully detect objects in varying conditions amidst background clutter and mild occlusion. In evaluating object detection approaches, several important methodological issues arise that have not been satisfactorily addressed in previous work. A secondary focus of this paper is to highlight these issues and to develop rigorous evaluation standards for the object detection problem. A critical evaluation of our approach under the proposed standards is presented.

Algorithms↗
PubMed2003

Environmentally mediated synergy between perception and behaviour in mobile robots.

The notion that behaviour influences perception seems self-evident, but the mechanism of their interaction is not known. Perception and behaviour are usually considered to be separate processes. In this view, perceptual learning constructs compact representations of sensory events, reflecting their statistical properties, independently of behavioural relevance. Behavioural learning, however, forms associations between perception and action, organized by reinforcement, without regard for the construction of perception. It is generally assumed that the interaction between these two processes is internal to the agent, and can be explained solely in terms of the neuronal substrate. Here we show, instead, that perception and behaviour can interact synergistically via the environment. Using simulated and real mobile robots, we demonstrate that perceptual learning directly supports behavioural learning and so promotes a progressive structuring of behaviour. This structuring leads to a systematic bias in input sampling, which directly affects the organization of the perceptual system. This external, environmentally mediated feedback matches the perceptual system to the emerging behavioural structure, so that the behaviour is stabilized.

Adaptation, Physiological↗
PubMed2004

Associative change in the representations acquired during conditional discriminations: further analysis of the nature of conditional learning.

Three experiments with rats investigated how the associative strengths of the representations that underlie conditional learning change when they are conditioned in compound. The results of each experiment suggest that the representation whose associative strength is most discrepant from the asymptote supported by the outcome of the trial undergoes the greatest change in associative strength. These results parallel those from simple Pavlovian conditioning (e.g., R. A. Rescorla, 2000). are inconsistent with unique-cue and configural accounts of conditional learning, and support a connectionist analysis of learning in which a "winner-takes-all" rule applies to the hidden units that can be activated and acquire associative strength at a given point in time.

Analysis of Variance↗
PubMed1990

Representational guidance of action production in observational learning: a causal analysis.

This experiment tested the hypothesis that the number of model presentations and verbal coding of modeled actions affect reproduction accuracy through their effect on cognitive representation. Subjects viewed a complex action pattern either two or eight times with or without verbal coding to highlight the dynamic structure of the component actions and their temporal sequencing. They then received, in order, a recognition test and a pictorial-arrangement test to assess the accuracy of their cognitive representations of the modeled actions. Subsequently, all subjects were tested for their ability to reproduce the action pattern from memory. Results showed that increased exposure to modeled actions enhanced the accuracy of both the cognitive representation and the behavioral reproduction. Verbal coding also increased cognitive and reproduction accuracy, but only when combined with multiple opportunities to observe the modeled actions. A causal analysis confirmed that the effects of multiple exposures and verbal coding were entirely mediated by changes produced in the accuracy of cognitive representation.

Journal Article↗
PubMed2001

Neuronal representations of stimulus associations develop in the temporal lobe during learning.

Visual stimuli that are frequently seen together become associated in long-term memory, such that the sight of one stimulus readily brings to mind the thought or image of the other. It has been hypothesized that acquisition of such long-term associative memories proceeds via the strengthening of connections between neurons representing the associated stimuli, such that a neuron initially responding only to one stimulus of an associated pair eventually comes to respond to both. Consistent with this hypothesis, studies have demonstrated that individual neurons in the primate inferior temporal cortex tend to exhibit similar responses to pairs of visual stimuli that have become behaviorally associated. In the present study, we investigated the role of these areas in the formation of conditional visual associations by monitoring the responses of individual neurons during the learning of new stimulus pairs. We found that many neurons in both area TE and perirhinal cortex came to elicit more similar neuronal responses to paired stimuli as learning proceeded. Moreover, these neuronal response changes were learning-dependent and proceeded with an average time course that paralleled learning. This experience-dependent plasticity of sensory representations in the cerebral cortex may underlie the learning of associations between objects.

Animals↗
PubMed1998

Three-Dimensional Representations of Contour Maps

Contour map learning may require three-dimensional representation of the area depicted by a contour map. The purpose of the study was to test the hypothesis that participants created three-dimensional mental representations of contour maps when asked to generate a cross-section profile of the map terrain from one point on the contour map to another. Gender differences in cross-section performance were also investigated, and it was hypothesized that males would perform better than females in a contour map cross-section test. Participants studied a series of contour maps or landsurface maps (three-dimensional drawings of an area), and answered two cross-section questions per map. Following the cross-section test, participants were given an incidental recognition test for the previously studied maps, either in a contour map format or a landsurface map format. Males recognized the landsurface maps corresponding to the contour map cross-section questions answered correctly better than contour map cross-section questions answered incorrectly, whereas females did not. This finding suggested that males, but not females, formed three-dimensional representations of the contour maps. There were no gender differences in performance on the contour map cross-section test, but males achieved higher recognition scores than females when the cross-section stimuli were contour maps. It was concluded that multiple spatial and verbal processing strategies can be used successfully to solve a contour map cross-section test, but that three-dimensional spatial processing may be more efficient than other forms of processing for long-term memory of contour maps. Copyright 1998 Academic Press.

Journal Article↗
PubMed1998

Learning-induced expansion of cortical maps--what happens to adjacent cortical representations?

The present study was designed to investigate the effects of learning-dependent enlargement of cortical representation of a row of vibrissae upon the appearance of adjacent cortical representations. We have found previously that three sessions of classical conditioning, during which stimulation of row B of vibrissae is paired with a tail shock, result in an increase of cortical representation of the trained row, as visualized with 2-deoxyglucose (2DG) metabolic functional mapping. In the present experiment, after the training in which row B was stimulated, we mapped with 2DG the cortical representations of rows A and C, not stimulated during the training. We found that these representations do not differ from normal. Because of expansion of cortical representation of the trained row, the overlap between representations of neighboring rows of vibrissae became greater and the two maps co-existed within the same cortical space.

Animals↗
PubMedDate not supplied

Development of neural mechanisms for machine learning.

The goal of this work is to develop a humanoid robot's perceptual mechanisms through the use of learning aids. We describe methods to enable learning on a humanoid robot using learning aids such as books, drawing materials, boards, educational videos or other children toys. Visual properties of objects are learned and inserted into a recognition scheme, which is then applied to acquire new object representations - we propose learning through developmental stages. Inspired in infant development, we will also boost the robot's perceptual capabilities by having a human caregiver performing educational and play activities with the robot (such as drawing, painting or playing with a toy train on a railway). We describe original algorithms to extract meaningful percepts from such learning experiments. Experimental evaluation of the algorithms corroborates the theoretical framework.

Algorithms↗
PubMed2004

Neural correlates of learned song in the avian forebrain: simultaneous representation of self and others.

Songbirds are extraordinary vocalists and sensitive listeners, singing to communicate identity, engage other birds in acoustical combat, and attract mates. These processes involve auditory plasticity in that birds rapidly learn to discriminate novel from familiar songs. Songbirds also are one of the few non-human animals that use auditory feedback to learn their vocalizations, thus auditory -- vocal interactions are likely to be important to vocal learning. Recent advances strengthen the connection between song recognition and processing of birdsong in the auditory telencephalon. New insights also have emerged into the mechanisms underlying the 'gating' of auditory responses and the emergence of highly selective responses, two processes that could facilitate auditory feedback important to song learning.

Animals↗
PubMed2003

For real! XCS with continuous-valued inputs.

Many real-world problems are not conveniently expressed using the ternary representation typically used by Learning Classifier Systems and for such problems an interval-based representation is preferable. We analyse two interval-based representations recently proposed for XCS, together with their associated operators and find evidence of considerable representational and operator bias. We propose a new interval-based representation that is more straightforward than the previous ones and analyse its bias. The representations presented and their analysis are also applicable to other Learning Classifier System architectures. We discuss limitations of the real multiplexer problem, a benchmark problem used for Learning Classifier Systems that have a continuous-valued representation, and propose a new test problem, the checkerboard problem, that matches many classes of real-world problem more closely than the real multiplexer. Representations and operators are compared using both the real multiplexer and checkerboard problems and we find that representational, operator and sampling bias all affect the performance of XCS in continuous-valued environments.

Classification↗