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 1,549 records · Page 86Linked to original sources
PubMed2003

Clustering ensembles of neural network models.

We show that large ensembles of (neural network) models, obtained e.g. in bootstrapping or sampling from (Bayesian) probability distributions, can be effectively summarized by a relatively small number of representative models. In some cases this summary may even yield better function estimates. We present a method to find representative models through clustering based on the models' outputs on a data set. We apply the method on an ensemble of neural network models obtained from bootstrapping on the Boston housing data, and use the results to discuss bootstrapping in terms of bias and variance. A parallel application is the prediction of newspaper sales, where we learn a series of parallel tasks. The results indicate that it is not necessary to store all samples in the ensembles: a small number of representative models generally matches, or even surpasses, the performance of the full ensemble. The clustered representation of the ensemble obtained thus is much better suitable for qualitative analysis, and will be shown to yield new insights into the data.

Algorithms↗
PubMed2004

Neuronal activity in the rodent dorsal striatum in sequential navigation: separation of spatial and reward responses on the multiple T task.

The striatum plays an important role in "habitual" learning and memory and has been hypothesized to implement a reinforcement-learning algorithm to select actions to perform given the current sensory input. Many experimental approaches to striatal activity have made use of temporally structured tasks, which imply that the striatal representation is temporal. To test this assumption, we recorded neurons in the dorsal striatum of rats running a sequential navigation task: the multiple T maze. Rats navigated a sequence of four T maze turns to receive food rewards delivered in two locations. The responses of neurons that fired phasically were examined. Task-responsive phasic neurons were active as rats ran on the maze (maze-responsive) or during reward receipt (reward-responsive). Neither maze- nor reward-responsive neurons encoded simple motor commands: maze-responses were not well correlated with the shape of the rat's path and most reward-responsive neurons did not fire at similar rates at both food-delivery sites. Maze-responsive neurons were active at one or more locations on the maze, but these responses did not cluster at spatial landmarks such as turns. Across sessions the activity of maze-responsive neurons was highly correlated when rats ran the same maze. Maze-responses encoded the location of the rat on the maze and imply a spatial representation in the striatum in a task with prominent spatial demands. Maze-responsive and reward-responsive neurons were two separate populations, suggesting a divergence in striatal information processing of navigation and reward.

Algorithms↗
PubMed2003

Kernel hierarchical gene clustering from microarray expression data.

MOTIVATION: Unsupervised analysis of microarray gene expression data attempts to find biologically significant patterns within a given collection of expression measurements. For example, hierarchical clustering can be applied to expression profiles of genes across multiple experiments, identifying groups of genes that share similar expression profiles. Previous work using the support vector machine supervised learning algorithm with microarray data suggests that higher-order features, such as pairwise and tertiary correlations across multiple experiments, may provide significant benefit in learning to recognize classes of co-expressed genes. RESULTS: We describe a generalization of the hierarchical clustering algorithm that efficiently incorporates these higher-order features by using a kernel function to map the data into a high-dimensional feature space. We then evaluate the utility of the kernel hierarchical clustering algorithm using both internal and external validation. The experiments demonstrate that the kernel representation itself is insufficient to provide improved clustering performance. We conclude that mapping gene expression data into a high-dimensional feature space is only a good idea when combined with a learning algorithm, such as the support vector machine that does not suffer from the curse of dimensionality. AVAILABILITY: Supplementary data at www.cs.columbia.edu/compbio/hiclust. Software source code available by request.

Algorithms↗
PubMed1990

Discriminative properties of morphine that modulate associations between tastes and lithium chloride.

Wistar rats learned to withhold consumption of a target solution when morphine preceded presentation of the target solution and lithium chloride (LiCl) and to consume the same target solution when saline preceded the presentation of the solution. After this serial feature discrimination training, morphine did not block the formation of a Pavlovian association between saccharin and LiCl but did suppress consumption of familiar tap water. After Pavlovian conditioning, morphine blocked the formation of an association between saccharin and LiCl but did not suppress consumption of a familiar tap water solution. The roles of morphine and saline can be interchanged. It appears that the morphine discriminative stimulus is calling up a representation of neither the conditioned stimulus nor the unconditional stimulus alone, but rather a modified representation of some aspect of their association.

Animals↗
PubMed1998

The diabolo classifier

We present a new classification architecture based on autoassociative neural networks that are used to learn discriminant models of each class. The proposed architecture has several interesting properties with respect to other model-based classifiers like nearest-neighbors or radial basis functions: it has a low computational complexity and uses a compact distributed representation of the models. The classifier is also well suited for the incorporation of a priori knowledge by means of a problem-specific distance measure. In particular, we will show that tangent distance (Simard, Le Cun, & Denker, 1993) can be used to achieve transformation invariance during learning and recognition. We demonstrate the application of this classifier to optical character recognition, where it has achieved state-of-the-art results on several reference databases. Relations to other models, in particular those based on principal component analysis, are also discussed.

Journal Article↗
PubMedDate not supplied

The ontogeny of phonological categories and the primacy of lexical learning in linguistic development.

In this paper, we draw on recent developments in several areas of cognitive science that suggest that the lexicon is at the core of grammatical generalizations at several different levels of representation. Evidence comes from many sources, including recent studies on language processing in adults and on language acquisition in children. Phonological behavior is influenced very early by pattern frequency in the lexicon of the ambient language, and we propose that phonological acquisition might provide the initial bootstrapping into grammatical generalization in general. The phonological categories over which pattern frequencies are calculated, however, are neither transparently available in the audiovisual signal nor deterministically fixed by the physiological and perceptual capacities of the species. Therefore, we need several age-appropriate models of how the lexicon can influence a child's interactions with the ambient language over the course of phonological acquisition.

Child↗
PubMed2005

Chemosensory processing in the fruit fly, Drosophila melanogaster: generalization of a feeding response reveals overlapping odour representations.

Insects are capable of detecting, and discriminating between, a very large number of odours. The biological relevance of many of those odours, particularly those related to food, must first be learned. Given that the number of sensory receptors and antennal lobe (AL) glomeruli is limited relative to the number of odours that must be detectable, this ability implies that the olfactory system makes use of a combinatorial coding scheme whereby each sensory cell or AL projection neuron can participate in coding for several different odours. An important step in understanding this coding scheme is to behaviourally quantify the degree to which sets of odours are discriminable. Here we evaluate odour discriminability in the fruit fly, Drosophila melanogaster, by first conditioning individual flies to not respond to any of several odorants using a nonassociative conditioning protocol (habituation). We show that flies habituate unconditioned leg movement responses to both mechanosensory and olfactory stimulation over 25 unreinforced trials. Habituation is retained for at least 2 h and is subject to dishabituation. Finally, we test the degree to which the conditioned response generalizes to other odorants based on molecular features of the odorants (e.g. carbon chain length and the presence of a target functional group). These tests reveal predictable generalization gradients across these molecular features. These data substantiate the claim that these features are relevant coding dimensions in the fruit fly olfactory system, as has been shown for other insect and vertebrate species.

Acyclic Monoterpenes↗
PubMed2005

Dietary restraint and US devaluation predict evaluative learning.

Previous research has indicated that flavor-flavor learning is impaired in restrained eaters. In Experiment 1 we sought to extend this finding using a larger sample and a more comprehensive assessment of dietary behavior. Participants (N=90, including 30 current dieters) sampled three novel flavors (CSs), each on 10 separate occasions, in a randomized order. Each flavor was paired with chocolate (US) either 10%, 50%, or 90% of the time. We then assessed liking for the three CSs and asked participants to complete the DEBQ-restraint and TFEQ-disinhibition sub-scales. After these CS-US parings, restrained eaters tended to prefer the 10% paired flavor whereas unrestrained eaters tended to prefer the 90% paired flavor. Differential CS liking was not evident in dieters and it was not predicted by disinhibition. Using a similar methodology, in Experiment 2 (N=76) we assessed evaluative change following picture-sweet pairings. Relative to the other CSs, the restrained eaters reported a greater increase in their liking for the 10% paired CS and the unrestrained eaters reported a greater increase in their liking for the 90% paired CS. We also discovered that evaluative change is related to the level of US devaluation that takes place during conditioning. Evidence that a sweet US can bring about a decrease in liking has not been reported previously. One interpretation is that negative beliefs and attitudes can contaminate the representation of the US during training.

Adult↗
PubMed2003

Encoding predicted outcome and acquired value in orbitofrontal cortex during cue sampling depends upon input from basolateral amygdala.

Certain goal-directed behaviors depend critically upon interactions between orbitofrontal cortex (OFC) and basolateral amygdala (ABL). Here we describe direct neurophysiological evidence of this cooperative function. We recorded from OFC in intact and ABL-lesioned rats learning odor discrimination problems. As rats learned these problems, we found that lesioned rats exhibited marked changes in the information represented in OFC during odor cue sampling. Lesioned rats had fewer cue-selective neurons in OFC after learning; the cue-selective population in lesioned rats did not include neurons that were also responsive in anticipation of the predicted outcome; and the cue-activated representations that remained in lesioned rats were less associative and more often bound to cue identity. The results provide a neural substrate for representing acquired value and features of the predicted outcome during cue sampling, disruption of which could account for deficits in goal-directed behavior after damage to this system.

Amygdala↗
PubMed1997

The effects of disorientation on visual landmark control of head direction cell orientation.

Head direction (HD) and place cells were recorded in rats that had previously exhibited significant acquisition deficits on a radial arm maze task following disorientation treatment. In this study we determined whether this behavioral impairment was associated with a lack of landmark stimulus control over the preferred orientations of HD and place cells. Neurons were recorded as animals retrieved food pellets in a cylindrical apparatus containing a single cue card. Some of these HD cells were also recorded while animals explored an eight-arm radial maze in a similar cue-controlled environment. The stimulus control of the landmarks in each environment was assessed by rotating the landmark and examining the subsequent preferred orientations of HD and place cells. Animals underwent disorientation treatment before and after each recording session. Despite this disorientation, rotation of the cue card in the cylindrical apparatus resulted in a corresponding shift in the preferred orientations of HD and place cells in 13 of 15 and 7 of 7 recording sessions, respectively. On the radial arm maze, rotation of the landmark cue was associated with a corresponding shift in the HD cell's preferred orientation in 7 of 9 sessions. These results suggest that a visual landmark's stimulus control may not require a learned association between that landmark and an animal's stable experience in an environment. Furthermore, instability in the HD cell system is unlikely to account for the impaired performance of the disoriented animals in the radial arm maze. Rather, these impairments may be due to the animal's inability to utilize stable representations of the environment provided by HD and place cells.

Animals↗
PubMed1996

Rapid visual learning in neurones of the primate temporal visual cortex.

The human visual system can learn to recognize visual stimuli rapidly. For example, humans can accurately reconstruct meaningful objects out of fragmentary evidence, once they have seen the same object in its unambiguous form. The anterior temporal cortical areas of macaques contain some neurones with invariant visual responses which appear to provide a representation of complex patterns and objects, such as faces. Remarkably, these neurones show an enhancement of response after brief (e.g. 5 s) exposure to the unambiguous stimulus, an effect that appears to reflect the neural basis of the rapid perceptual learning seen in humans.

Animals↗
PubMed1997

Putting faces to names.

An experiment is reported which investigated participants' ability to remember a person's face when they were presented with the person's name. During the learning phase, participants were shown 18 unfamiliar faces together with a name and occupation. At test, participants were presented with a name and were asked to indicate the face and occupation that had been presented with that name at learning. Results showed that participants' ability to remember the face was contingent upon their ability to remember the occupation that had been presented with the name. When participants were presented with a face and were asked to remember the name, performance was also contingent upon correctly remembering the associated occupation, consistent with the findings of McWeeny, Young, Hay & Ellis (1987). No such contingencies were apparent when participants were given an occupation and were asked to remember the associated name and face; participants frequently remembered the name but not the face, or the face without the name. These results are consistent with the serial access model of person identification proposed by Bruce & Young (1986), and with more recent developments proposed by Valentine, Brédart, Lawson & Ward (1991) and Craigie & Hanley (1993) in which there are no direct links between the representation of a person's name in memory and visual information about their facial appearance. The mnemonic strategies that were used by the participants during the learning phase of the experiment were also examined. These strategies help explain those few occasions on which participants appear to be able to link names to faces without identity-specific semantic information.

Adolescent↗
PubMed1990

Castration anxiety and pregenital development: a hypothesis on the protophallus.

This paper discusses some aspects of the mental position of the penis and their relevance to castration anxiety. We know that the genital occupies an uncertain position in the representation of the body, due to the fact that the penis escapes the synthetic functions of the ego. Since we assume that this is caused partly by the processes occurring when a child learns to stand and walk, it is possible to establish a link between the phallus of the phallic phase and the imaginary phallus of the pregenital period at the time when an upright body posture is adopted. I postulate that, at the time when a child learns to stand and walk, a protophallus is created in the form of an auxiliary leg, which performs a complex function and, as a mental fetish, forms the basis of the imaginary phallus. A bodily precursor of castration anxiety can be seen in the process described. This protophallus, the imaginary phallus and the phallus of the phallic phase are later all absorbed into the psychical representation of the penis and determine the mental image in the long term. In a discussion of the foundation of this proposition in psychoanalytic theory, I point to the heuristic value of this concept for a number of fields of knowledge and fields of application.

Anxiety, Castration↗
PubMed2006

Learning to attend: modeling the shaping of selectivity in infero-temporal cortex in a categorization task.

Recent experiments on behaving monkeys have shown that learning a visual categorization task makes the neurons in infero-temporal cortex (ITC) more selective to the task-relevant features of the stimuli (Sigala and Logothetis in Nature 415 318-320, 2002). We hypothesize that such a selectivity modulation emerges from the interaction between ITC and other cortical area, presumably the prefrontal cortex (PFC), where the previously learned stimulus categories are encoded. We propose a biologically inspired model of excitatory and inhibitory spiking neurons with plastic synapses, modified according to a reward based Hebbian learning rule, to explain the experimental results and test the validity of our hypothesis. We assume that the ITC neurons, receiving feature selective inputs, form stronger connections with the category specific neurons to which they are consistently associated in rewarded trials. After learning, the top-down influence of PFC neurons enhances the selectivity of the ITC neurons encoding the behaviorally relevant features of the stimuli, as observed in the experiments. We conclude that the perceptual representation in visual areas like ITC can be strongly affected by the interaction with other areas which are devoted to higher cognitive functions.

Algorithms↗
PubMed1995

Modulation of muscle responses evoked by transcranial magnetic stimulation during the acquisition of new fine motor skills.

1. We used transcranial magnetic stimulation (TMS) to study the role of plastic changes of the human motor system in the acquisition of new fine motor skills. We mapped the cortical motor areas targeting the contralateral long finger flexor and extensor muscles in subjects learning a one-handed, five-finger exercise on the piano. In a second experiment, we studied the different effects of mental and physical practice of the same five-finger exercise on the modulation of the cortical motor areas targeting muscles involved in the task. 2. Over the course of 5 days, as subjects learned the one-handed, five-finger exercise through daily 2-h manual practice sessions, the cortical motor areas targeting the long finger flexor and extensor muscles enlarged, and their activation threshold decreased. Such changes were limited to the cortical representation of the hand used in the exercise. No changes of cortical motor outputs occurred in control subjects who underwent daily TMS mapping but did not practice on the piano at all (control group 1). 3. We studied the effect of increased hand use without specific skill learning in subjects who played the piano at will for 2 h each day using only the right hand but who were not taught the five-finger exercise (control group 2) and who did not practice any specific task. In these control subjects, the changes in cortical motor outputs were similar but significantly less prominent than in those occurring in the test subjects, who learned the new skill.(ABSTRACT TRUNCATED AT 250 WORDS)

Adult↗
PubMed2006

Maternal secure-base scripts and children's attachment security in an adopted sample.

Studies of families with adopted children are of special interest to attachment theorists because they afford opportunities to probe assumptions of attachment theory with regard to the developmental timing of interactions necessary to form primary attachments and also with regard to effects of shared genes on child attachment quality. In Bowlby's model, attachment-relevant behaviors and interactions are observable from the moment of birth, but for adoptive families, these interactions cannot begin until the child enters the family, sometimes several months or even years post-partum. Furthermore, because adoptive parents and adopted children do not usually share genes by common descent, any correspondence between attachment representations of the parent and secure base behavior of the child must arise as a consequence of dyadic interaction histories. The objectives of this study were to evaluate whether the child's age at the time of adoption or at the time of attachment assessment predicted child attachment security in adoptive families and also whether the adoptive mother's internal attachment representation predicted the child's attachment security. The participants were 106 mother - child dyads selected from the 406 adoptions carried out through the Lisbon Department of Adoption Services over a period of 3 years. The Attachment Behavior Q-Set (AQS; Waters, 1995) was used to assess secure base behavior and an attachment script representation task was used to assess the maternal attachment representations. Neither child's age at the time of adoption, nor age of the child at assessment significantly predicted the AQS security score; however, scores reflecting the presence and quality of maternal secure base scripts did predict AQS security. These findings support the notion that the transmission of attachment security across generations involves mutual exchanges and learning by the child and that the exchanges leading to secure attachment need not begin at birth. These results complement the findings and conceptual arguments offered by Bowlby and Ainsworth concerning the critical influence of maternal representations of attachment to the quality of attachment security in children.

Adoption↗
PubMed1997

Peripheral and cerebral asymmetries in the rat.

Rats learn a novel foraging pattern better with their right-side whiskers than with their left-side whiskers. They also learn better with the left cerebral hemisphere than with the right hemisphere. Rotating an already learned maze relative to the external environment most strongly reduces right-whisker performance; starting an already learned maze at a different location most strongly reduces left-whisker performance. These results suggest that the right-periphery-left-hemisphere system accesses a map-like representation of the foraging problem, whereas the left-periphery-right-hemisphere system accesses a rote path. Thus, as in humans, functional asymmetries in rats can be elicited by both peripheral and cortical manipulation, and each hemisphere makes qualitatively distinct contributions to a complex natural behavior.

Animals↗
PubMed1999

[The specific cognitive stile of pain patients supports chronification of pain].

BACKGROUND: In a cognitive perspective, chronic pain comprises at least three dimensions: First it is possible to study the relevance of pain related attitudes, beliefs and coping cognitions for the chronification of pain. Second psychological processes of learning and memory processes can be analysed. Third we can investigate uncontrolled cognitions in chronic pain patients. AIM OF THE STUDY: The first part of the present paper deals with representations of pain events in autobiographical memory. In the second part a hindsight bias experiment is used as a prototype of altered information processing in the context of chronic pain. STUDY 1: In study one recollection of pain related events, pain experience and the sensory recalling of pain occurrences were sampled in 20 chronic pain patients, 17 psychiatric patients and 38 healthy controls. Pain patients showed a specific kind of pain related memory which had no parallel among psychiatric patients. Based on learning theory the significance of a pain related memory for chronification is discussed. STUDY 2: In the second study 18 pain patients, 13 psychiatric patients and 18 healthy controls were tested with a hindsight bias experiment. The hindsight-effect was observed in the usual extend in the student control group, but was significantly greater in the pain group and absent in the psychiatric sample. In addition to this global finding, multinomial modeling revealed group differences in specific model parameters. Basic units of information processing interact with the chronification of pain. This method of analysis thus proved as a promising tool for the assessment of cognitive aspects of clinical disorders.

English Abstract↗