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 433 records · Page 24Linked to original sources
PubMed2006

Evidentiality in language and cognition.

What is the relation between language and thought? Specifically, how do linguistic and conceptual representations make contact during language learning? This paper addresses these questions by investigating the acquisition of evidentiality (the linguistic encoding of information source) and its relation to children's evidential reasoning. Previous studies have hypothesized that the acquisition of evidentiality is complicated by the subtleness and abstractness of the underlying concepts; other studies have suggested that learning a language which systematically (e.g. grammatically) marks evidential categories might serve as a pacesetter for early reasoning about sources of information. We conducted experimental studies with children learning Korean (a language with evidential morphology) and English (a language without grammaticalized evidentiality) in order to test these hypotheses. Our experiments compared 3- and 4-year-old Korean children's knowledge of the semantics and discourse functions of evidential morphemes to their (non-linguistic) ability to recognize and report different types of evidential sources. They also compared Korean children's source monitoring abilities to the source monitoring abilities of English-speaking children of the same age. We found that Korean-speaking children have considerable success in producing evidential morphology but their comprehension of such morphology is very fragile. Nevertheless, young Korean speakers are able to reason successfully about sources of information in non-linguistic tasks; furthermore, their performance in these tasks is similar to that of English-speaking peers. These results support the conclusion that the acquisition of evidential expressions poses considerable problems for learners; however, these problems are not (necessarily) conceptual in nature. Our data also suggest that, contrary to relativistic expectations, children's ability to reason about sources of information proceeds along similar lines in diverse language-learning populations and is not tied to the acquisition of the linguistic markers of evidentiality in the exposure language. We discuss implications of our findings for the relationship between linguistic and conceptual representations during development.

Asian People↗
PubMed1993

Issues in the acquisition of the Sesotho tonal system.

This paper examines the acquisition of the grammatical tone system of Sesotho, a southern Bantu language where tone sandhi is rich, and where surface and underlying representations are often quite distinct. Results of the longitudinal case study show that rule-assigned tone on subject markers is generally marked appropriately by age two. In contrast, underlying tonal representations on verb roots are learned gradually over time, showing an early Default High tone pattern. The study also finds that, while some tone sandhi rules are in the process of being acquired between 2;6 and 3;0, problems in the mapping between tonal representations and segments persist. The paper raises methodological and theoretical issues not only for the acquisition of tonal systems, but for the acquisition of phonology in general.

Acoustic Stimulation↗
PubMed2004

Effects of training on neuronal activity and interactions in primary and higher visual cortices in the alert cat.

The effects of behavioral training on early visual representations have been elusive when assessed with firing rates. Learning-induced changes in performance, however, suggest that representations should encompass early cortical stages. Here, we address the question of whether training-induced effects are pertinent to neuronal activity outside the task proper, which is a requirement if subsequent perceptional processes should profit from training. To search for a neuronal signature of training effects beyond firing rates, we measured local field potentials, multiunit and isolated spike activity during passive viewing of previously learned stimulus response associations (S+ and S-) in areas 17/18 and 21a of two alert cats. Evoked potential responses as well as gamma oscillations even during the first 200 msec were found to be stronger for S+ in both areas. Most importantly, the later parts of the response (>200 msec) not only exhibit a highly significant difference in coherent gamma oscillations for S+ and S- both within and across areas, but are also characterized by a pronounced preference in firing rate for S+ in area 21a, whereas primary cortex shows a nonsignificant trend for weaker spike responses. From these results, we conclude that training-induced plasticity occurs in adult visual cortex for behaviorally relevant stimuli by changing primarily the temporal structure of neuronal activity at early stages of cortical processing, whereas later stages of cortical processing express the increased coherence of their input in elevated firing rates.

Action Potentials↗
PubMed2006

Perceptual learning in speech: stability over time.

Perceptual representations of phonemes are flexible and adapt rapidly to accommodate idiosyncratic articulation in the speech of a particular talker. This letter addresses whether such adjustments remain stable over time and under exposure to other talkers. During exposure to a story, listeners learned to interpret an ambiguous sound as [f] or [s]. Perceptual adjustments measured after 12 h were as robust as those measured immediately after learning. Equivalent effects were found when listeners heard speech from other talkers in the 12 h interval, and when they had the opportunity to consolidate learning during sleep.

Attention↗
PubMed1993

Hippocampal mediation of stimulus representation: a computational theory.

The authors propose a computational theory of the hippocampal region's function in mediating stimulus representations. The theory assumes that the hippocampal region develops new stimulus representations that enhance the discriminability of differentially predictive cues while compressing the representation of redundant cues. Other brain regions, including cerebral and cerebellar cortices, are presumed to use these hippocampal representations to recode their own stimulus representations. In the absence of an intact hippocampal region, the theory implies that other brain regions will attempt to learn associations using previously established fixed representations. Instantiated as a connectionist network model, the theory provides a simple and unified interpretation of the functional role of the hippocampal region in a wide range of conditioning paradigms, including stimulus discrimination, reversal learning, stimulus generalization, latent inhibition, sensory preconditioning, and contextual sensitivity. The theory makes novel predictions regarding the effects of hippocampal lesions on easy-hard transfer and compound preexposure. Several prior qualitative characterizations of hippocampal function--including stimulus selection, chunking, cue configuration, and contextual coding--are identified as task-specific special cases derivable from this more general theory. The theory suggests that a profitable direction for future empirical and theoretical research will be the study of learning tasks in which both intact and lesioned animals exhibit similar initial learning behaviors but differ on subsequent transfer and generalization tasks.

Animals↗
PubMed1991

Topological maps of protein sequences.

A new method based on neural networks to cluster proteins into families is described. The network is trained with the Kohonen unsupervised learning algorithm, using matrix pattern representations of the protein sequences as inputs. The components (x, y) of these 20 x 20 matrix patterns are the normalized frequencies of all pairs xy of amino acids in each sequence. We investigate the influence of different learning parameters in the final topological maps obtained with a learning set of ten proteins belonging to three established families. In all cases, except in those where the synaptic vectors remains nearly unchanged during learning, the ten proteins are correctly classified into the expected families. The classification by the trained network of mutated or incomplete sequences of the learned proteins is also analysed. The neural network gives a correct classification for a sequence mutated in 21.5% +/- 7% of its amino acids and for fragments representing 7.5% +/- 3% of the original sequence. Similar results were obtained with a learning set of 32 proteins belonging to 15 families. These results show that a neural network can be trained following the Kohonen algorithm to obtain topological maps of protein sequences, where related proteins are finally associated to the same winner neuron or to neighboring ones, and that the trained network can be applied to rapidly classify new sequences. This approach opens new possibilities to find rapid and efficient algorithms to organize and search for homologies in the whole protein database.

Amino Acid Sequence↗
PubMed1997

Factor analysis using delta-rule wake-sleep learning.

We describe a linear network that models correlations between real-valued visible variables using one or more real-valued hidden variables-a factor analysis model. This model can be seen as a linear version of the Helmholtz machine, and its parameters can be learned using the wake-sleep method, in which learning of the primary generative model is assisted by a recognition model, whose role is to fill in the values of hidden variables based on the values of visible variables. The generative and recognition models are jointly learned in wake and sleep phases, using just the delta rule. This learning procedure is comparable in simplicity to Hebbian learning, which produces a somewhat different representation of correlations in terms of principal components. We argue that the simplicity of wake-sleep learning makes factor analysis a plausible alternative to Hebbian learning as a model of activity-dependent cortical plasticity.

Factor Analysis, Statistical↗
PubMed1997

Evolutionary induction of sparse neural trees

This paper is concerned with the automatic induction of parsimonious neural networks. In contrast to other program induction situations, network induction entails parametric learning as well as structural adaptation. We present a novel representation scheme called neural trees that allows efficient learning of both network architectures and parameters by genetic search. A hybrid evolutionary method is developed for neural tree induction that combines genetic programming and the breeder genetic algorithm under the unified framework of the minimum description length principle. The method is successfully applied to the induction of higher order neural trees while still keeping the resulting structures sparse to ensure good generalization performance. Empirical results are provided on two chaotic time series prediction problems of practical interest.

Journal Article↗
PubMed2005

[The effects of associative strength between lexical and conceptual representations on word processing in second language learners].

The purpose of this study was to examine how the learning experience affects associative strength between lexical representations of first and second language (L1, and L2) and conceptual representation, and word processing. The critical experimental manipulations were L2 proficiency, word frequency, and matching direction. The subjects were advanced learners and beginners. Association strength was assessed by measuring reaction time and error rate in translation recognition tasks with three sets of English (L2) and Japanese (L1) words that were high or low frequency, and corresponding pictures. The results showed that (1)the two associations involving L2 words were stronger when L2 word frequency was high, (2)both L2-L1 and L2-Picture association were strengthened with L2 proficiency. However, (3)there was no difference between L2-L1 and L2-Picture conditions. Moreover, (4) the association between L1 words and pictures were strong regardless of L2 proficiency or word frequency. These results suggest that L2 learning experience strengthens both associations involving L2 words.

Association↗
PubMed2006

Changes in auditory cortex parallel rapid perceptual learning.

Learning perceptual skills is characterized by rapid improvements in performance within the first hour of training (fast perceptual learning) followed by more gradual improvements that take place over several daily practice sessions (slow perceptual learning). Although it is widely accepted that slow perceptual learning is accompanied by enhanced stimulus representation in sensory cortices, there is considerable controversy about the neural substrates underlying early and rapid improvements in learning perceptual skills. Here we measured event-related brain potentials while listeners were presented with 2 phonetically different vowels. Listeners' ability to identify both vowels improved gradually during the first hour of testing and was paralleled by enhancements in an early evoked response ( approximately 130 ms) localized in the right auditory cortex and a late evoked response ( approximately 340 ms) localized in the right anterior superior temporal gyrus and/or inferior prefrontal cortex. These neuroplastic changes depended on listeners' attention and were preserved only if practice was continued; familiarity with the task structure (procedural learning) was not sufficient. We propose that the early increases in cortical responsiveness reflect goal-directed changes in the tuning properties of auditory neurons involved in parsing concurrent speech signals. Importantly, the neuroplastic changes occurred rapidly, demonstrating the flexibility of human speech segregation mechanisms.

Acoustic Stimulation↗
PubMed2006

Images, frames, and connectionist hierarchies.

The representation of hierarchically structured knowledge in systems using distributed patterns of activity is an abiding concern for the connectionist solution of cognitively rich problems. Here, we use statistical unsupervised learning to consider semantic aspects of structured knowledge representation. We meld unsupervised learning notions formulated for multilinear models with tensor product ideas for representing rich information. We apply the model to images of faces.

Classification↗
PubMed2026

ChromBERT-tools: a versatile toolkit for context-specific regulatory representations of transcription regulators across different cell types.

SUMMARY: Representations that encode the genome-wide regulatory behavior of transcription regulators provide a foundation for flexible transcription modeling and in silico regulatory analysis. Existing regulator representations are commonly derived from gene co-expression, motif annotations, or static protein features, which capture useful but limited aspects of regulator identity but do not directly model how regulators participate in region-specific regulatory programs across the genome. ChromBERT addresses this gap by learning context-aware regulatory representations from large-scale ChIP-seq data. However, routine bioinformatics applications require lightweight, accessible, and modular tools for generating, adapting, and interpreting these representations in user-defined biological contexts. Here, we present ChromBERT-tools, a user-oriented toolkit built upon ChromBERT that converts its regulatory representation framework into practical workflows for customizable analysis across cellular contexts. ChromBERT-tools provides command-line interfaces and Python APIs organized into three functional layers: representation generation, predictive modeling, and regulatory interpretation. The representation generation layer produces representations of genomic regions and transcription regulators. The predictive modeling layer fine-tunes ChromBERT for genome-wide regulatory activity prediction through classification or regression tasks, with optimized implementation to reduce running time and computational resource requirements. The regulatory interpretation layer supports inference of the context-specific roles of cis-regulatory elements and transcription regulators. These modules can be used independently or integrated into end-to-end workflows, enabling flexible analyses across diverse datasets. ChromBERT-tools lowers the barrier to applying context-specific regulatory representations in routine genomic analyses. AVAILABILITY AND IMPLEMENTATION: ChromBERT-tools is freely available at https://github.com/TongjiZhanglab/ChromBERT-tools, with documentation at https://chrombert-tools.readthedocs.io/en/latest/. A frozen archival snapshot is available on Zenodo under DOI: 10.5281/zenodo.20094206.

Software↗
PubMed2006

Chunking in task sequences modulates task inhibition.

In a study of the formation of representations of task sequences and its influence on task inhibition, participants first performed tasks in a predictable sequence (e.g., ABACBC) and then performed the tasks in a random sequence. Half of the participants were explicitly instructed about the predictable sequence, whereas the other participants did not receive these instructions. Task-sequence learning was inferred from shorter reaction times (RTs) in predictable relative to random sequences. Persisting inhibition of competing tasks was indicated by increased RTs in n- 2 task repetitions (e.g., ABA) compared with n- 2 nonrepetitions (e.g., CBA). The results show task-sequence learning for both groups. However, task inhibition was reduced in predictable relative to random sequences among instructed-learning participants who formed an explicit representation of the task sequence, whereas sequence learning and task inhibition were independent in the noninstructed group. We hypothesize that the explicit instructions led to chunking of the task sequence, and that n- 2 repetitions served as chunk points (ABA-CBC), so that within-chunk facilitation modulated the inhibition effect.

Adult↗
PubMed2000

The effects of category use on learned categories.

Frequently, people learn to classify instances of a concept and later learn additional information about the concept. What is the effect of this later learning on the original classification? In five experiments, this issue was investigated with a common classification paradigm in which symptom sets were classified into disease categories. After learning to classify these sets, the subjects learned to use the category to decide what treatment should be given for a symptom set. The symptoms that were important for the treatments were later classified by disease more accurately and were generated earlier from the disease name. However, this effect occurred only if the category representation was activated during the learning of the treatments. Thus, later learning about a particular use of the concept can sometimes affect the original classification.

Adult↗
PubMed2004

A novel algorithm for scalable and accurate Bayesian network learning.

Bayesian Networks (BN) is a knowledge representation formalism that has been proven to be valuable in biomedicine for constructing decision support systems and for generating causal hypotheses from data. Given the emergence of datasets in medicine and biology with thousands of variables and that current algorithms do not scale more than a few hundred variables in practical domains, new efficient and accurate algorithms are needed to learn high quality BNs from data. We present a new algorithm called Max-Min Hill-Climbing (MMHC) that builds upon and improves the Sparse Candidate (SC) algorithm; a state-of-the-art algorithm that scales up to datasets involving hundreds of variables provided the generating networks are sparse. Compared to the SC, on a number of datasets from medicine and biology, (a) MMHC discovers BNs that are structurally closer to the data-generating BN, (b) the discovered networks are more probable given the data, (c) MMHC is computationally more efficient and scalable than SC, and (d) the generating networks are not required to be uniformly sparse nor is the user of MMHC required to guess correctly the network connectivity

Algorithms↗
PubMed1993

The perception of form and motion.

Although form and motion are two distinct aspects of visual processing, they do not start as separate entities in the visual system. Early analyses extract discontinuities in various image attributes and these can trace the outline of a form. When displaced, the same image features can give rise to impressions of motion. Recent work has overturned many of the assumptions about the contributions of different stimulus attributes to motion processing, and reorganized the classification of motion systems. The results have revealed unexpected interactions between attention and motion. Paralleling this research is work on the early stages of form learning and on the nature of the stored representations.

Attention↗
PubMed1997

Supervised Learning Extensions to the CLAM Network.

The contextual layered associative memory (CLAM) has been developed as a self-generating structure which implements a probabilistic encoding scheme. The training algorithms are geared towards the unsupervised generation of a layerable associative mapping ([Thacker and Mayhew, 1989]). We show here that the resulting structure will support layers which can be trained to produce outputs that approximate conditional probabilities of classification. Unsupervised and supervised learning algorithms operate independently permitting the unsupervised representational layer to be developed before supervision is available. The system thus supports learning which is inherently more flexible than conventional node labelling schemes. Copyright 1997 Elsevier Science Ltd. All Rights Reserved.

Journal Article↗