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 523 records · Page 29Linked to original sources
PubMed2000

Visual management of large scale data mining projects.

This paper describes a unified framework for visualizing the preparations for, and results of, hundreds of machine learning experiments. These experiments were designed to improve the accuracy of enzyme functional predictions from sequence, and in many cases were successful. Our system provides graphical user interfaces for defining and exploring training datasets and various representational alternatives, for inspecting the hypotheses induced by various types of learning algorithms, for visualizing the global results, and for inspecting in detail results for specific training sets (functions) and examples (proteins). The visualization tools serve as a navigational aid through a large amount of sequence data and induced knowledge. They provided significant help in understanding both the significance and the underlying biological explanations of our successes and failures. Using these visualizations it was possible to efficiently identify weaknesses of the modular sequence representations and induction algorithms which suggest better learning strategies. The context in which our data mining visualization toolkit was developed was the problem of accurately predicting enzyme function from protein sequence data. Previous work demonstrated that approximately 6% of enzyme protein sequences are likely to be assigned incorrect functions on the basis of sequence similarity alone. In order to test the hypothesis that more detailed sequence analysis using machine learning techniques and modular domain representations could address many of these failures, we designed a series of more than 250 experiments using information-theoretic decision tree induction and naive Bayesian learning on local sequence domain representations of problematic enzyme function classes. In more than half of these cases, our methods were able to perfectly discriminate among various possible functions of similar sequences. We developed and tested our visualization techniques on this application.

Alcohol Dehydrogenase↗
PubMed2002

[Interactive effect of MK-801 and pre-training on spatial learning in rats].

In experiment 1, the effect of an NMDA receptor antagonist, MK-801, on the formation of the spatial representation was investigated. The administration of 0.1 mg/kg of MK-801 induced learning deficits in rats (n = 10) with the Morris watermaze task. However, when rats (n = 10) were pre-trained in the same task without drug treatment, and then trained in the same task in a different environment under the influence of the same amount of the drug, their performance was not impaired. The result suggests that rats treated with MK-801 can acquire a spatial representation of their environment, and that the impairment shown by the drug-treated rats without pre-training may be due to the impairment in the learning of the problem-solving strategy required for the watermaze place task. Experiment 2 examined this possibility. Rats (n = 10) were pre-trained with a visual cue discrimination task without drug treatment and then trained in the place task with MK-801 (0.1 mg/kg) treatment. They did not show impairment in the place task, indicating that rats treated with MK-801 can learn a new problem-solving strategy. Thus the learning deficits of MK-801-treated rats that have not been pre-trained do not seem to be due to impaired acquisition of the spatial representation or of the learning of strategy required to solve the watermaze place task.

Animals↗
PubMedDate not supplied

Early lexical development in a self-organizing neural network.

In this paper we present a self-organizing neural network model of early lexical development called DevLex. The network consists of two self-organizing maps (a growing semantic map and a growing phonological map) that are connected via associative links trained by Hebbian learning. The model captures a number of important phenomena that occur in early lexical acquisition by children, as it allows for the representation of a dynamically changing linguistic environment in language learning. In our simulations, DevLex develops topographically organized representations for linguistic categories over time, models lexical confusion as a function of word density and semantic similarity, and shows age-of-acquisition effects in the course of learning a growing lexicon. These results match up with patterns from empirical research on lexical development, and have significant implications for models of language acquisition based on self-organizing neural networks.

Artificial Intelligence↗
PubMed2005

Dietary learning in humans: directions for future research.

There is every indication that most of our flavor preferences and dietary behaviors are learned. Despite this, we know very little about the underlying mechanisms. This paper considers reasons why this might be the case. In addition to considering particular methodological issues and the potential relevance of a 'critical developmental period', emphasis is placed on the need to resolve whether learning results from an implicit or an explicit process. Addressing this issue has important implications for the way that studies should be designed. It also leads to one of two diametrically opposite conclusions. Either behavior is governed by a rather rare form of automatic and involuntary associative learning, or otherwise, it should be regarded as non-automatic and subject to attentional and other constraints associated with most other forms of learning in humans. This latter proposition invites speculation that learning might also be governed by more complex representations (beliefs and attitudes) associated with the foods and flavors that are presented in learning studies. More generally, an analysis of this kind is important because it has the potential to explain differences that might underpin particular aberrant eating habits.

Animals↗
PubMed1994

The locus of knowledge effects in concept learning.

Three experiments investigated how knowledge influences concept formation and representation in a standard concept acquisition task. The primary comparison was among arbitrary concepts, which had meaningless features; meaningful concepts, which had meaningful features from different domains; and integrated concepts, which had meaningful features interconnected by common knowledge. Experiment 1 found that learning was superior for the integrated concepts but that there was little difference as a function of feature meaningfulness. Experiment 2 suggested that the integrated Ss were learning to form a knowledge-based schema as their concept representation because they did not distinguish the typicality of features that differed in frequency. Experiment 3 introduced a category whose features were from the same domain but were not otherwise related. This concept was as difficult to learn and use as the meaningful concepts were. These comparisons help specify the ways in which knowledge does and does not influence concept formation.

Concept Formation↗
PubMed2003

Implicit scene learning is viewpoint dependent.

When novel scenes are encoded, the representations of scene layout are generally viewpoint specific. Past studies of scene recognition have typically required subjects to explicitly study and encode novel scenes, but in everyday visual experience, it is possible that much scene learning occurs incidentally. Here, we examine whether implicitly encoded scene layouts are also viewpoint dependent. We used the contextual cuing paradigm, in which search for a target is facilitated by implicitly learned associations between target locations and novel spatial contexts (Chun & Jiang, 1998). This task was extended to naturalistic search arrays with apparent depth. To test viewpoint dependence, the viewpoint of the scenes was varied from training to testing. Contextual cuing and, hence, scene context learning decreased as the angular rotation from training viewpoint increased. This finding suggests that implicitly acquired representations of scene layout are viewpoint dependent.

Child↗
PubMed1996

Using concept maps on the World-Wide Web to access a curriculum database for problem-based learning.

Development of medical school curriculum databases continues to be challenging. Representation of the instructional unit is becoming increasingly difficult due to characteristics of the problem-based learning (PBL) curricula. Curriculum databases may be used to store materials for the PBL curricula, and also to provide a delivery mechanism for those materials. However, in order to take advantage of the curriculum database as a tool for PBL, methods for accessing the curriculum database that are better suited to the information needs of students, faculty, and administrators must be developed. Concept maps are directed graph representations of conceptual relationships, and may be used to represent the content of a curriculum database. In this paper, we describe a Web application that uses Java-based concept maps was the user interface to a curriculum database.

Computer Communication Networks↗
PubMed2005

Theta rhythm of navigation: link between path integration and landmark navigation, episodic and semantic memory.

Five key topics have been reverberating in hippocampal-entorhinal cortex (EC) research over the past five decades: episodic and semantic memory, path integration ("dead reckoning") and landmark ("map") navigation, and theta oscillation. We suggest that the systematic relations between single cell discharge and the activity of neuronal ensembles reflected in local field theta oscillations provide a useful insight into the relationship among these terms. In rats trained to run in direction-guided (1-dimensional) tasks, hippocampal cell assemblies discharge sequentially, with different assemblies active on opposite runs, i.e., place cells are unidirectional. Such tasks do not require map representation and are formally identical with learning sequentially occurring items in an episode. Hebbian plasticity, acting within the temporal window of the theta cycle, converts the travel distances into synaptic strengths between the sequentially activated and unidirectionally connected assemblies. In contrast, place representations by hippocampal neurons in 2-dimensional environments are typically omnidirectional, characteristic of a map. Generation of a map requires exploration, essentially a dead reckoning behavior. We suggest that omnidirectional navigation through the same places (junctions) during exploration gives rise to omnidirectional place cells and, consequently, maps free of temporal context. Analogously, multiple crossings of common junction(s) of episodes convert the common junction(s) into context-free or semantic memory. Theta oscillation can hence be conceived as the navigation rhythm through both physical and mnemonic space, facilitating the formation of maps and episodic/semantic memories.

Animals↗
PubMed2002

Cognitive mechanisms of transitive inference.

We examined how the brain organizes interrelated facts during learning and how the facts are subsequently manipulated in a transitive inference (TI) paradigm (e.g., if A<B and B<C, then A<C). This task determined features such as learned facts and behavioral goals, but the learned facts could be organized in any of several ways. For example, if one learns a list by operating on paired items, the pairs may be stored individually as separate facts and reaction time (RT) should decrease with learning. Alternatively, the pairs may be stored as a single, unified list, which may yield a different RT pattern. We characterized RT patterns that occurred as participants learned, by trial and error, the predetermined order of 11 shapes. The task goal was to choose the shape occurring closer to the end of the list, and feedback about correctness was provided during this phase. RT increased even as its variance decreased during learning, suggesting that the learnt knowledge became progressively unified into a single representation, requiring more time to manipulate as participants acquired relational knowledge. After learning, non-adjacent (NA) list items were presented to examine how participants reasoned in a TI task. The task goal also required choosing from each presented pair the item occurring closer to the list end, but without feedback. Participants could solve the TI problems by applying formal logic to the previously learnt pairs of adjacent items; alternatively, they could manipulate a single, unified representation of the list. Shorter RT occurred for NA pairs having more intervening items, supporting the hypothesis that humans employ unified mental representations during TI. The response pattern does not support mental logic solutions of applying inference rules sequentially, which would predict longer RT with more intervening items. We conclude that the brain organizes information in such a way that reflects the relations among the items, even if the facts were learned in an arbitrary order, and that this representation is subsequently used to make inferences.

Adolescent↗
PubMed1997

What can we learn from the morphology of Hebrew? A masked-priming investigation of morphological representation.

All Hebrew words are composed of 2 interwoven morphemes: a triconsonantal root and a phonological word pattern. the lexical representations of these morphemic units were examined using masked priming. When primes and targets shared an identical word pattern, neither lexical decision nor naming of targets was facilitated. In contrast root primes facilitated both lexical decisions and naming of target words that were derived from these roots. This priming effect proved to be independent of meaning similarity because no priming effects were found when primes and targets were semantically but not morphologically related. These results suggest that Hebrew roots are lexical units whereas word patterns are not. A working model of lexical organization in Hebrew is offered on the basis of these results.

Decision Making↗
PubMed2001

Teaching Bayesian reasoning in less than two hours.

The authors present and test a new method of teaching Bayesian reasoning, something about which previous teaching studies reported little success. Based on G. Gigerenzer and U. Hoffrage's (1995) ecological framework, the authors wrote a computerized tutorial program to train people to construct frequency representations (representation training) rather than to insert probabilities into Bayes's rule (rule training). Bayesian computations are simpler to perform with natural frequencies than with probabilities, and there are evolutionary reasons for assuming that cognitive algorithms have been developed to deal with natural frequencies. In 2 studies, the authors compared representation training with rule training; the criteria were an immediate learning effect, transfer to new problems, and long-term temporal stability. Rule training was as good in transfer as representation training, but representation training had a higher immediate learning effect and greater temporal stability.

Algorithms↗
PubMed2004

An efficient and effective region-based image retrieval framework.

An image retrieval framework that integrates efficient region-based representation in terms of storage and complexity and effective on-line learning capability is proposed. The framework consists of methods for region-based image representation and comparison, indexing using modified inverted files, relevance feedback, and learning region weighting. By exploiting a vector quantization method, both compact and sparse (vector) region-based image representations are achieved. Using the compact representation, an indexing scheme similar to the inverted file technology and an image similarity measure based on Earth Mover's Distance are presented. Moreover, the vector representation facilitates a weighted query point movement algorithm and the compact representation enables a classification-based algorithm for relevance feedback. Based on users' feedback information, a region weighting strategy is also introduced to optimally weight the regions and enable the system to self-improve. Experimental results on a database of 10,000 general-purposed images demonstrate the efficiency and effectiveness of the proposed framework.

Abstracting and Indexing↗
PubMed2003

Assembling and encoding word representations: fMRI subsequent memory effects implicate a role for phonological control.

Novel word learning is central to the flexibility inherent in the human language capacity. Word learning may partially depend on long-term memory formation during the assembly of phonological representations from orthographic inputs. In the present study, event-related functional magnetic resonance imaging (fMRI) examined the contributions of phonological control-a component of the verbal working memory system-to phonological assembly and word learning. Subjects were scanned while making syllable decisions about visually presented familiar (English) and novel (pseudo-English and Foreign) words, a task that required retrieval and analysis of existing phonological codes or the assembly and analysis of novel representations. Results revealed that left inferior prefrontal cortex (LIPC) and bilateral parietal cortices were differentially engaged during the processing of novel words, suggesting that this circuit is recruited during phonological assembly. A subsequent memory analysis that examined the relation between fMRI signal and the subject's ability to later remember the words (a measure of effective memory formation) revealed that the magnitude of activation in LIPC, bilateral superior parietal, and left inferior parietal cortices was positively correlated with later memory. Moreover, although the magnitude of the subsequent memory effect in parietal cortex was not significantly affected by word type, this effect was greater in posterior LIPC for novel (pseudo-English) than for familiar (English) words. In the course of subserving the assembly of novel word representations, the phonological (articulatory) control component of the phonological system appears to play a central role in the encoding of novel words into long-term memory.

Adolescent↗
PubMed2005

Learning and transfer of an ipsilateral coordination task: evidence for a dual-layer movement representation.

The present study addressed the nature of the memory representation for interlimb coordination tasks. For this purpose, the acquisition of a multifrequency (2:1) task with the ipsilateral limbs and transfer to the ipsilateral and contralateral body side was examined. In particular, subjects practiced a 2:1 coordination pattern whereby the right arm moved twice as fast as the right leg, or vice versa. Subsequently, they transferred the practiced 2:1 task to three different conditions: (1) the converse partner (i.e., the slow-moving limb had to move fast, and vice versa) at the ipsilateral body side, and (2) the identical and (3) converse 2:1 pattern at the contralateral body side. Findings revealed positive transfer of the identical and converse 2:1 pattern to the contralateral body side. However, no transfer of the learned pattern to its converse partner at the same body side was revealed. We propose a new memory representation model for coordination patterns, composed of an effector-independent and effector-specific component (dual-layer model). It is hypothesized that the general movement goal (i.e., moving one limb twice as fast as the other) constitutes the abstract, higher-level representation that may account for positive contralateral transfer. Conversely, the effector-specific component contains task-specific lower-level muscle synergies that are acquired through practice, prohibiting positive transfer when shifting task allocation within the same effectors. These findings are consistent with recent neuroscientific evidence for neuroplastic changes in distributed brain areas.

Adult↗
PubMed2004

Mental models and meaningful learning.

If you understand something, you can use the information you have acquired to solve problems to which that knowledge is relevant. Meaningful learning is learning with understanding. Achieving meaningful learning begins with the building of correct, appropriate mental models, or representations, of the knowledge being acquired. The next step is learning to use the available mental models to solve problems. In many of the biomedical sciences, this means being able to either calculate something, predict the responses of the system, or explain the responses of the system. Since only the learner can do the learning, the only possible role for the teacher is to help the learner to learn. This means creating an active learning environment in which the learner can acquire the needed information, continually test the mental models being built, and correct or refine those models as needed. In an active learning environment, students are given ample opportunities to learn to solve problems. If the goal of the course is the achievement of meaningful learning, it is essential that the students then be assessed to determined whether they have reached that goal.

Animals↗
PubMed2005

Problem-based learning and learning approach: is there a relationship?

AIM: To assess the influence of a graduate-entry PBL curriculum on individual learning style; and to investigate the relationship between learning style, academic achievement and clinical reasoning skill. METHOD: Subjects were first-year medical students completed the Study Process Questionnaire at the commencement, and again, at the end of the academic year when they also completed the Diagnostic Thinking Inventory, a measure of clinical reasoning skill. Subjects were classified on the basis of their predominant learning approach, and this was correlated with examination results and DTI score. RESULTS: There was a net shift in predominant learning approach away from deep learning towards a more surface approach over the period of the study, as well as a significant decrease in deep-learning scores. There was a statistically significant association between deep learning score and clinical reasoning skill as shown by total DTI score as well as on the structure of knowledge subscale. No correlation was found between learning approach and examination results. CONCLUSION: Although these results suggest that a deep learning approach may be beneficial in the development of clinical reasoning skill through its potential to enhance the development of knowledge representations, the substantial shift towards a surface learning approach brings into question previous conclusions that PBL curricula foster a deep approach to learning, and suggests that other factors, such as work load may be more determinants of learning approach than curriculum type. Taken together, these findings emphasise the context-dependent nature of learning approach as well as the importance of assessment as a driver of student learning and strongly suggest that further work to determine precisely the factors which influence learning approach in medical students is urgently needed.

Adult↗
PubMed1997

Generative models for discovering sparse distributed representations.

We describe a hierarchical, generative model that can be viewed as a nonlinear generalization of factor analysis and can be implemented in a neural network. The model uses bottom-up, top-down and lateral connections to perform Bayesian perceptual inference correctly. Once perceptual inference has been performed the connection strengths can be updated using a very simple learning rule that only requires locally available information. We demonstrate that the network learns to extract sparse, distributed, hierarchical representations.

Algorithms↗
PubMed2000

Representing the task in Bayesian reasoning: comment on Lovett and Schunn (1999).

The RCCL model (M. C. Lovett & C. D. Schunn, 1999) produces predictions that are non-novel or that do not truly spring from its principles. However, it offers the valuable insight that learning processes may affect the selection of both representations and strategies within those representations, and points the way to possible theoretical progress on implicit and explicit control. The authors' account of base-rate neglect under direct experience is compared with RCCL, and it is concluded that learning-based models allow for tests that are not fostered by representation-based models.

Bayes Theorem↗