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 775 records · Page 43Linked to original sources
PubMed1999

What are the computations of the cerebellum, the basal ganglia and the cerebral cortex?

The classical notion that the cerebellum and the basal ganglia are dedicated to motor control is under dispute given increasing evidence of their involvement in non-motor functions. Is it then impossible to characterize the functions of the cerebellum, the basal ganglia and the cerebral cortex in a simplistic manner? This paper presents a novel view that their computational roles can be characterized not by asking what are the "goals" of their computation, such as motor or sensory, but by asking what are the "methods" of their computation, specifically, their learning algorithms. There is currently enough anatomical, physiological, and theoretical evidence to support the hypotheses that the cerebellum is a specialized organism for supervised learning, the basal ganglia are for reinforcement learning, and the cerebral cortex is for unsupervised learning.This paper investigates how the learning modules specialized for these three kinds of learning can be assembled into goal-oriented behaving systems. In general, supervised learning modules in the cerebellum can be utilized as "internal models" of the environment. Reinforcement learning modules in the basal ganglia enable action selection by an "evaluation" of environmental states. Unsupervised learning modules in the cerebral cortex can provide statistically efficient representation of the states of the environment and the behaving system. Two basic action selection architectures are shown, namely, reactive action selection and predictive action selection. They can be implemented within the anatomical constraint of the network linking these structures. Furthermore, the use of the cerebellar supervised learning modules for state estimation, behavioral simulation, and encapsulation of learned skill is considered. Finally, the usefulness of such theoretical frameworks in interpreting brain imaging data is demonstrated in the paradigm of procedural learning.

Journal Article↗
PubMed2003

Polynomial harmonic GMDH learning networks for time series modeling.

This paper presents a constructive approach to neural network modeling of polynomial harmonic functions. This is an approach to growing higher-order networks like these build by the multilayer GMDH algorithm using activation polynomials. Two contributions for enhancement of the neural network learning are offered: (1) extending the expressive power of the network representation with another compositional scheme for combining polynomial terms and harmonics obtained analytically from the data; (2) space improving the higher-order network performance with a backpropagation algorithm for further gradient descent learning of the weights, initialized by least squares fitting during the growing phase. Empirical results show that the polynomial harmonic version phGMDH outperforms the previous GMDH, a Neurofuzzy GMDH and traditional MLP neural networks on time series modeling tasks. Applying next backpropagation training helps to achieve superior polynomial network performances.

Algorithms↗
PubMed2001

Pharmacokinetic software for the health sciences: choosing the right package for teaching purposes.

Computer assisted learning has an important role in the teaching of pharmacokinetics to health sciences students because it transfers the emphasis from the purely mathematical domain to an 'experiential' domain in which graphical and symbolic representations of actions and their consequences form the major focus for learning. Basic pharmacokinetic concepts can be taught by experimenting with the interplay between dose and dosage interval with drug absorption (e.g. absorption rate, bioavailability), drug distribution (e.g. volume of distribution, protein binding) and drug elimination (e.g. clearance) on drug concentrations using library ('canned') pharmacokinetic models. Such 'what if' approaches are found in calculator-simulators such as PharmaCalc, Practical Pharmacokinetics and PK Solutions. Others such as SAAM II, ModelMaker, and Stella represent the 'systems dynamics' genre, which requires the user to conceptualise a problem and formulate the model on-screen using symbols, icons, and directional arrows. The choice of software should be determined by the aims of the subject/course, the experience and background of the students in pharmacokinetics, and institutional factors including price and networking capabilities of the package(s). Enhanced learning may result if the computer teaching of pharmacokinetics is supported by tutorials, especially where the techniques are applied to solving problems in which the link with healthcare practices is clearly established.

Computer Simulation↗
PubMed2005

Word learning by preschoolers with specific language impairment: effect of phonological or semantic cues.

PURPOSE: This study investigated whether phonological or semantic encoding cues promoted better word learning for children with specific language impairment (SLI) and whether this treatment differentially affected children with SLI and normal language (NL). METHOD: Twenty-four preschoolers ages 4;0 (years;months) to 5;11 with SLI and 24 age- and gender-matched children with NL participated. The between-group factor was language group (NL, SLI) and within-group factors were language modality (comprehension, recognition, production) and treatment condition (phonological, semantic). Word learning was assessed during fast mapping, word learning, and post-testing with trials to criterion calculated for the number of words learned. A drawing task assessed the change in semantic representation of words. RESULTS: The SLI group comprehended more words in the semantic condition and produced more words in the phonological condition, but the NL group performed similarly in both. The NL group required significantly fewer trials than the SLI group to comprehend words in the semantic and phonological conditions and to produce words in the semantic condition, but between-group differences for production were not significant for the phonological condition. CONCLUSIONS: The results suggest that preschoolers with SLI may benefit from cues that highlight the phonological or semantic properties of words but that different cues may aid different aspects of word learning.

Analysis of Variance↗
PubMed2003

A grounded theory of abstraction in artificial intelligence.

In artificial intelligence, abstraction is commonly used to account for the use of various levels of details in a given representation language or the ability to change from one level to another while preserving useful properties. Abstraction has been mainly studied in problem solving, theorem proving, knowledge representation (in particular for spatial and temporal reasoning) and machine learning. In such contexts, abstraction is defined as a mapping between formalisms that reduces the computational complexity of the task at stake. By analysing the notion of abstraction from an information quantity point of view, we pinpoint the differences and the complementary role of reformulation and abstraction in any representation change. We contribute to extending the existing semantic theories of abstraction to be grounded on perception, where the notion of information quantity is easier to characterize formally. In the author's view, abstraction is best represented using abstraction operators, as they provide semantics for classifying different abstractions and support the automation of representation changes. The usefulness of a grounded theory of abstraction in the cartography domain is illustrated. Finally, the importance of explicitly representing abstraction for designing more autonomous and adaptive systems is discussed.

Artificial Intelligence↗
PubMed1998

Using olfaction to study memory.

In a series of studies we have been exploring the role of hippocampal function in memory using the model system of olfactory-hippocampal pathways and odor learning in rats. Our experiments show that hippocampus itself is not essential to memory for single odors, but is critical for forming the representations of relations among odor memories, and for the expression of odor memory representations in novel situations. These studies that exploit the exceptional qualities of olfactory learning are helping to clarify the nature of higher order memory processes in all mammals, and extending to declarative memory in humans.

Animals↗
PubMed2005

The acquisition of face and person identity information following anterior temporal lobectomy.

Thirty unilateral anterior temporal lobectomy (ATL) subjects (15 right and 15 left) and 15 controls were presented a multitrial learning task in which unfamiliar faces were paired with biographical information (occupation, city location, and a person's name). Face recognition hits were similar between groups, but the right ATL group committed more false-positive errors to face foils. Both left and right ATL groups were impaired relative to controls in acquiring biographical information, but the deficit was more pronounced for the left ATL group. Recall levels also varied for the different types of biographical information; occupation was most commonly recalled followed by city name and person name. In addition, city and person name recall was more likely when occupation was also recalled. Overall, recall of biographical information was positively correlated with clinical measures of anterograde episodic memory. Findings are discussed in terms of the role of the temporal lobe and associative learning ability in the successful acquisition of new face semantic (biographical) representations.

Adult↗
PubMed1994

Probabilistic analysis of human supervised learning and classification.

Probabilistic classification techniques based on Bayesian decision theory are used to analyze human supervised learning and classification. The procedure rests on the assumption that human classification behaviour is based on internal feature states which can be linked to physical feature vectors (corresponding to the system input). In the present approach, this relationship is modeled in terms of additive stochastic error signals. The corresponding random variables describe the additional degrees of bias and variance introduced by the (perceptual) process of internal feature measurement. Estimates of internal feature states are obtained by least-squares minimization. Structure and dimensionality of the resulting internal representation are displayed by plotting the configuration of internal class means, or virtual prototypes. Their temporal evolution reflects the dynamic properties of the learning process. The use of the procedure is demonstrated by analyzing the results of two experiments. First, it is shown that Minimum Distance Classifiers, such as used by Caelli, Rentschler and Scheidler [(1987) Biological Cybernetics, 57, 233-240], are suboptimal in predicting human performance. Second, it is found that extrafoveal learning is much slower than foveal learning and that extrafoveal pattern representations are severely distorted. The latter distortions reveal the existence of limitations for the generalization of supervised learning over space.

Adult↗
PubMed2002

Model-based processing scheme for quantitative 4-D cardiac MRI analysis.

In this paper, we present an integrated model-based processing scheme for cardiac magnetic resonance imaging (MRI), embedded in an interactive computing environment suitable for quantitative cardiac analysis, which provides a set of functions for the extraction, modeling, and visualization of cardiac shape and deformation. The methods apply four-dimensional (4-D) processing (three spatial and one temporal) to multiphase multislice MRI acquisitions and produce a continuous 4-D model of the myocardial surface deformation. The model is used to measure diagnostically useful parameters, such as wall motion, myocardial thicking, and myocardial mass measurements. The proposed model-based shape extraction method has the advantage of integrating local information into an overall representation and produces a robust description of cardiac cavities. A learning segmentation process that incorporates a generating-shrinking neural network is combined with a spatiotemporal parametric modeling method through functional basis decomposition. A multiscale approach is adopted, which uses at each step a coarse-scale model defined at the previous step in order to constrain the boundary detection. The representation accuracy starts from a coarse but smooth estimation of the approximate cardiac shape and is gradually increased to the desired detail. The main advantages of the proposed methods are efficiency, lack of uncertainty about convergence, and robustness to image artifacts. Experimental results obtained from application to clinical multislice multiphase MRI examinations of normal volunteers and patients with medical record of myocardial infarction were satisfactory in terms of accuracy and robustness.

Heart↗
PubMed1999

Learning to recognize objects.

Evidence from neurophysiological and psychological studies is coming together to shed light on how we represent and recognize objects. This review describes evidence supporting two major hypotheses: the first is that objects are represented in a mosaic-like form in which objects are encoded by combinations of complex, reusable features, rather than two-dimensional templates, or three-dimensional models. The second hypothesis is that transform-invariant representations of objects are learnt through experience, and that this learning is affected by the temporal sequence in which different views of the objects are seen, as well as by their physical appearance.

Journal Article↗
PubMed2006

Are there distinct neural representations of object and limb dynamics?

In recent studies of human motor learning, subjects learned to move the arm while grasping a robotic device that applied novel patterns of forces to the hand. Here, we examined the generality of force field learning. We tested the idea that contextual cues associated with grasping a novel object promote the acquisition and use of a distinct internal model, associated with that object. Subjects learned to produce point-to-point arm movements to targets in a horizontal plane while grasping a robotic linkage that applied either a velocity-dependent counter-clockwise or clockwise force field to the hand. Following adaptation, subjects let go of the robot and were asked to generate the same movements in free space. Small but reliable after-effects were observed during the first eight movements in free space, however, these after-effects were significantly smaller than those observed for control subjects who moved the robot in a null field. No reduction in retention was observed when subjects subsequently returned to the force field after moving in free space. In contrast, controls who reached with the robot in a NF showed much poorer retention when returning to a force field. These findings are consistent with the idea that contextual cues associated with grasping a novel object may promote the acquisition of a distinct internal model of the dynamics of the object, separate from internal models used to control limb dynamics alone.

Adolescent↗
PubMed2003

The role of domain knowledge in automating medical text report classification.

OBJECTIVE: To analyze the effect of expert knowledge on the inductive learning process in creating classifiers for medical text reports. DESIGN: The authors converted medical text reports to a structured form through natural language processing. They then inductively created classifiers for medical text reports using varying degrees and types of expert knowledge and different inductive learning algorithms. The authors measured performance of the different classifiers as well as the costs to induce classifiers and acquire expert knowledge. MEASUREMENTS: The measurements used were classifier performance, training-set size efficiency, and classifier creation cost. RESULTS: Expert knowledge was shown to be the most significant factor affecting inductive learning performance, outweighing differences in learning algorithms. The use of expert knowledge can affect comparisons between learning algorithms. This expert knowledge may be obtained and represented separately as knowledge about the clinical task or about the data representation used. The benefit of the expert knowledge is more than that of inductive learning itself, with less cost to obtain. CONCLUSION: For medical text report classification, expert knowledge acquisition is more significant to performance and more cost-effective to obtain than knowledge discovery. Building classifiers should therefore focus more on acquiring knowledge from experts than trying to learn this knowledge inductively.

Algorithms↗
PubMed1997

Some functions of primary auditory cortex in learning and memory formation.

In the primary auditory field AI of gerbil auditory cortex, aversive tone conditioning paradigms reshaped frequency receptive fields of single units and also changed the spatial representation of tones in fluoro-2-deoxyglucose (FDG) experiments. As another aspect of learning-induced plasticity in gerbil AI, antibodies against the immediate early gene product c-Fos identified an unusual spatial pattern of neurons in terms of a "macrocolumn." The pattern resulted from repeated short exposure of the animals to a tone in a new environment. The search for transmitters that may mediate this gene activation is carried out by microdialysis through chronically implanted probes in auditory cortex. So far, dopamine transmission was found to reflect specific aspects of auditory learning in cortex. The results suggest that spectral features of sounds as well as aspects of learned behavioral meaning of the sounds may be represented in AI.

Animals↗
PubMed2003

Prior knowledge and subtyping effects in children's category learning.

Two experiments examined how 5- and 10-year-old children revise their category representations when exposed to exemplars that are congruent or incongruent with existing knowledge. During training children were presented with exemplars containing features that were congruent or incongruent with children's social stereotypes together with a stereotype-neutral feature. In the knowledge-subtyping condition this neutral feature predicted the stereotype-congruence of the other features. In the knowledge-standard condition the neutral feature was uncorrelated with stereotype-congruence. At test children made judgements about feature co-occurrence within the learned category. In each experiment these judgements were influenced by both stereotypical beliefs and exemplar observation. Stereotypical beliefs, however, had a greater influence on co-occurrence judgements in the knowledge-subtyping than in the standard conditions. In Experiment 2 these effects were shown to generalize to judgements about features that were not presented during training. These results challenge current models of knowledge-based categorization by showing that exemplar structure determines whether novel exemplar features are incorporated into category representations.

Child↗
PubMed2006

Hippocampal lesion prevents spatial relational learning in adult macaque monkeys.

The role of the hippocampus in spatial learning and memory has been extensively studied in rodents. Comparable studies in nonhuman primates, however, are few, and findings are often contradictory. This may be attributable to the failure to distinguish between allocentric and egocentric spatial representations in experimental designs. For this experiment, six adult monkeys received bilateral hippocampal ibotenic acid lesions, and six control subjects underwent sham surgery. Freely moving monkeys then foraged for food located in two arrays of three distinct locations among 18 locations distributed in an open-field arena. Multiple goals and four pseudorandomly chosen entrance points precluded the monkeys' ability to rely on an egocentric strategy to identify food locations. Monkeys were tested in two conditions. First, local visual cues marked the food locations. Second, no local cues marked the food locations, so that monkeys had to rely on an allocentric (spatial relational) representation of the environment to discriminate these locations. Both hippocampal-lesioned and control monkeys discriminated the food locations in the presence of local cues. However, in the absence of local cues, control subjects discriminated the food locations, whereas hippocampal-lesioned monkeys were unable to do so. Interestingly, histological analysis of the brain of one control monkey whose behavior was identical to that of the experimentally lesioned animals revealed a bilateral ischemic lesion restricted to the hippocampus. These findings demonstrate that the adult monkey hippocampal formation is critical for the establishment or use of allocentric spatial representations and that selective damage of the hippocampus prevents spatial relational learning in adult nonhuman primates.

Animals↗
PubMed2005

Distinct basal ganglia territories are engaged in early and advanced motor sequence learning.

In this study, we used functional MRI (fMRI) at high field (3T) to track the time course of activation in the entire basal ganglia circuitry, as well as other motor-related structures, during the explicit learning of a sequence of finger movements over a month of training. Fourteen right-handed healthy volunteers had to practice 15 min daily a sequence of eight moves using the left hand. MRI sessions were performed on days 1, 14 and 28. In both putamen, activation decreased with practice in rostrodorsal (associative) regions. In contrast, there was a significant signal increase in more caudoventral (sensorimotor) regions of the putamen. Subsequent correlation analyses between signal variations and behavioral variables showed that the error rate (movement accuracy) was positively correlated with signal changes in areas activated during early learning, whereas reaction time (movement speed) was negatively correlated with signal changes in areas activated during advanced learning stages, including the sensorimotor putamen and globus pallidus. These results suggest the possibility that motor representations shift from the associative to the sensorimotor territories of the striato-pallidal complex during the explicit learning of motor sequences, suggesting that motor skills are stored in the sensorimotor territory of the basal ganglia that supports a speedy performance.

Adult↗
PubMed2005

The effect of semantic representation on toddlers' word retrieval.

PURPOSE: This study tested the hypothesis that depth of semantic representation influences toddlers' word retrieval. METHOD: Nineteen toddlers participated under 3 word learning conditions in this longitudinal study. Gestures cued attention to object shape (SHP) or function (FNC) in the experimental conditions. No semantic cue was provided under a control condition (CTL). Word learning conditions occurred on each of 3 days. On the 4th day, word retrieval was assessed across 3 levels of scaffolding (uncued picture naming, cued picture naming, picture recognition). Evidence of semantic representation was provided at fast and slow mapping intervals. RESULTS: Less scaffolding was necessary for word retrieval (uncued and cued naming) under experimental conditions than under the CTL condition. However, more SHP than FNC condition targets were retrieved for uncued picture naming. This latter difference may be related to the superior fast mapping of targets under the SHP condition. Toddlers stated object functions (slow mapping) comparably in the experimental conditions, but this was superior to CTL condition performance. CONCLUSIONS: Word retrieval is a continuous behavior that is positively influenced by semantic representation. Semantic knowledge of objects can be enriched by shape or function gestures, thereby improving toddlers' object word productions. Shape cues appear to be more effective for this purpose.

Analysis of Variance↗