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 1,225 records · Page 68Linked to original sources
PubMed2005

Acquiring long-term representations of visual classes following extensive extrastriate damage.

Different areas of human visual cortex are thought to play different roles in the learning of visual information: whereas in low/intermediate cortical areas, plasticity may be manifested by enhanced selectivity to learned visual features, in higher-level areas, plasticity may result in generalization and development of tolerance to degraded versions of the learned stimuli. The most effective tolerance to degraded information is presumably achieved in the case of cooperation between the different forms of plasticity. Whether this tolerance to degraded information also applies when the visual input is degraded as a result of a lesion to lower levels of the visual system remains an open question. To address this, we studied visual classification learning in a patient with an extensive bilateral lesion affecting intermediate/low-level visual areas but sparing higher-level areas. Despite difficulty in perceiving the stimuli, the patient learned to classify them, albeit not as quickly as control participants. Moreover, the patient's learning was maintained over the long term and was accompanied by improved discrimination of individual stimuli. These findings demonstrate that degraded output from lesioned, lower areas can be exploited in the service of a new visual task and the results likely implicate a combination of bottom-up and top-down processing during visual learning.

Agnosia↗
PubMed2003

Computational approaches to motor learning by imitation.

Movement imitation requires a complex set of mechanisms that map an observed movement of a teacher onto one's own movement apparatus. Relevant problems include movement recognition, pose estimation, pose tracking, body correspondence, coordinate transformation from external to egocentric space, matching of observed against previously learned movement, resolution of redundant degrees-of-freedom that are unconstrained by the observation, suitable movement representations for imitation, modularization of motor control, etc. All of these topics by themselves are active research problems in computational and neurobiological sciences, such that their combination into a complete imitation system remains a daunting undertaking-indeed, one could argue that we need to understand the complete perception-action loop. As a strategy to untangle the complexity of imitation, this paper will examine imitation purely from a computational point of view, i.e. we will review statistical and mathematical approaches that have been suggested for tackling parts of the imitation problem, and discuss their merits, disadvantages and underlying principles. Given the focus on action recognition of other contributions in this special issue, this paper will primarily emphasize the motor side of imitation, assuming that a perceptual system has already identified important features of a demonstrated movement and created their corresponding spatial information. Based on the formalization of motor control in terms of control policies and their associated performance criteria, useful taxonomies of imitation learning can be generated that clarify different approaches and future research directions.

Humans↗
PubMed2004

The vestibulo-ocular reflex as a model system for motor learning: what is the role of the cerebellum?

Motor systems are under a continuous adaptive process to maintain behavior throughout developmental changes and disease, a process called motor learning. Simple behaviors with easily measurable inputs and outputs are best suited to understand the neuronal signals that contribute to the required motor learning. Considering simple behaviors, the vestibulo-ocular reflex (VOR) allows quantification of its input and motor output and its neural circuitry is among the best documented. The main candidates for plastic change are the cerebellum and its target neurons in the brainstem. This review focuses on recent data regarding the involvement of the cerebellum in VOR motor learning. Learning can be divided into that acutely acquired over a period of hours and that chronically acquired over longer periods. Both acute and chronic learning have three phases named acquisition, consolidation, and retention. The cerebellar role in retention is disputed, but there is a consensus on the need of an intact cerebellum for acquisition. Data from neuronal recording, lesion studies and transgenic mouse experiments is complex but suggests that the signal representation in the cerebellum contains aspects of both motor output and sensory input. The cerebellum apparently uses different mechanisms for acute and chronic learning as well as for increases and decreases in VOR gain. Recent studies also suggest that the signal content in the cerebellum changes following learning and that the mechanisms used for chronic adaptation involve not only changes in a head velocity component but also in the efference copy of an eye movement command signal reaching Purkinje cells. This data leads to a new conceptual framework having implications for developing theories on the role of the cerebellum in motor learning and in the search for plastic elements within the VOR circuitry. For chronic learning we hypothesize that changes in the head velocity information traveling through the circuitry occur in parallel with changes in the integrator pathway and the efference copy pathway. We further propose that these changes are necessary to maintain the broadband characteristics of the learned behavior.

Animals↗
PubMed1993

Increasing the naming speed of poor readers: representations formed across repetitions.

In three experiments we examined the effect of repetition practice on the acquisition, retention, and generalization of children's skill in rapidly naming visually presented words. Experiment 1 showed that naming times decrease rapidly with practice. Retention of this newly acquired skill in rapid naming was a function of the degree of learning during training. "Overlearning" was necessary to prevent forgetting of the skill. Experiment 2 indicated that the rate at which the naming gains were acquired, and the amount of forgetting, was unrelated to the specific orthographic to phonological correspondences among the trained words. Experiment 3 suggested that even when the spelling/sound regularities were used to facilitate learning, there was no generalization of the naming time skill to new words that share the same spelling/sound relations. The results were discussed in terms of the nature of the representation that underlies rapid naming.

Child↗
PubMed2003

Learning-related changes in response patterns of prefrontal neurons during instrumental conditioning.

A crucial aspect of organizing goal-directed behavior is the ability to form neural representations of relationships between environmental stimuli, actions and reinforcement. Very little is known yet about the neural encoding of response-reward relationships, a process which is deemed essential for purposeful behavior. To investigate this, tetrode recordings were made in the medial prefrontal cortex (PFC) of rats performing a Go-NoGo task. After task acquisition, a subset of neurons showed a sustained change in firing during the rewarded action sequence that was triggered by a specific visual cue. When these changes were monitored in the course of learning, they were seen to develop in parallel with the behavioral learning curve and were highly sensitive to a switch in reward contingencies. These sustained changes correlated with the reward-associated action sequence, not with sensory or reward-predicting properties of the cue or individual motor acts per se. This novel type of neural plasticity may contribute to the formation of response-reinforcer associations and of behavioral strategies for guiding goal-directed action.

Action Potentials↗
PubMed2003

Category use and category learning.

Categorization models based on laboratory research focus on a narrower range of explanatory constructs than appears necessary for explaining the structure of natural categories. This mismatch is caused by the reliance on classification as the basis of laboratory studies. Category representations are formed in the process of interacting with category members. Thus, laboratory studies must explore a range of category uses. The authors review the effects of a variety of category uses on category learning. First, there is an extensive discussion contrasting classification with a predictive inference task that is formally equivalent to classification but leads to a very different pattern of learning. Then, research on the effects of problem solving, communication, and combining inference and classification is reviewed.

Classification↗
PubMed1995

Spatial- and locomotion-related neural representation in rat hippocampus following long-term survival from ischemia.

Spatial and locomotion-related behavioral correlates of hippocampal cell discharge were compared between ischemic and sham-control rats performing a spatial maze. Ischemic rats showed impaired choice accuracy during maze acquisition, but not during asymptote performance. Single-unit correlates during asymptote performance revealed enhanced spatial selectivity of CA2/3 complex-spike cells coincident with attenuated place-specific firing by hilar complex-spike or subicular cells. Responsivity to locomotion state by stratum granulosum interneurons was exaggerated, and locomotion-induced changes in firing of hilar and subicular interneurons was reduced. Ischemic rats showed recovered spatial learning abilities as evidenced by the fact that acquisition of the spatial task in a second environment was not impaired. Because representational reorganization was also observed in ischemic, maze-naive rats, brain injury per se appears to change information coding schemes.

Animals↗
PubMed2003

Take the best versus simultaneous feature matching: probabilistic inferences from memory and effects of representation format.

In 4 experiments, the tendency to use the simple heuristic Take The Best (TTB; G. Gigerenzer & D. Goldstein, 1996) was explored for probabilistic multiattribute inferences from memory. In a newly developed procedure, participants first learned attribute patterns that formed the basis for inferences in a second phase. A Bayesian method classified strategies as TTB, compensatory, or guessing. Experiment 1 had a high rate (64%) of participants classified as TTB users when inferences were made from memory. Experiment 2 showed that this was no mere materials effect. In Experiments 3 and 4, the authors examined effects of the representational format of the attribute information. Experiment 4 showed that the representational format may be an important moderating variable for strategy use.

Adult↗
PubMed2003

Severe cognitive and motor coordination deficits in tenascin-R-deficient mice.

The extracellular matrix molecule tenascin-R (TN-R), predominantly expressed in the central nervous system, has been implied in a variety of functions, e.g. during myelination, cerebellar neurite fasciculation and hippocampal long-term potentiation. In this study, we investigated in detail the impact of TN-R deficiency on the living animal by analyzing the behavior of TN-R-deficient mice. The general state, gross sensory functions, reflexes and motoric capabilities appeared normal. In contrast, motor coordination on the rota-rod was compromised in these mice, indicating a deficit in cerebellar functions. In the open field and the hole board, the mutants interact differently with their environment, probably due to differences in their exploratory behavior. TN-R-deficient mice were able to learn a reference memory task in the Morris water maze. In contrast to wild-type mice, the mutants displayed an alternative strategy; swimming around the pool using a stereotypical circling pattern, crossing all possible platform positions after relocation of the escape platform (reversal). These results, confirmed by relocating the platform in the center of the pool, suggest that TN-R-deficient mice may be impaired in constructing a goal-independent representation of space. In addition, a two-way active avoidance test (shuttle box) revealed a severe deficit in associative learning in TN-R-deficient mice. Our results support important functions of TN-R in vivo in the central nervous system, in particular in the cerebellum and the hippocampus.

Animals↗
PubMed2003

Stimulus representation in SOP: II. An application to inhibition of delay.

The componential extension of SOP accounts for conditioned response (CR) timing in Pavlovian conditioning by assuming that learning accrues with relative independence to stimulus elements that are differentially occasioned during the duration of the conditioned stimulus (CS). SOP, using a competitive learning rule and the assumption that temporal learning emerges via resolution of what is equivalent to an "AX+BX-" discrimination, predicts a progressive increase in the latency of the CR over training, or what Pavlov refer to as "inhibition of delay." Other componential models, which use noncompetitive learning rules, do not predict inhibition of delay. Either type of model makes the prediction indicated, independently of the length of the CS-unconditioned stimulus (US) interval. We report two experiments that demonstrated inhibition of delay when rabbits were trained with relatively long, but not with short, CS-US intervals. To account for this divergence, we assumed that the SOP stimulus trace involves two kinds of elements, some with a temporally distributed pattern of activity over the duration of the CS duration, and some with a randomly distributed pattern. This stimulus representation, not only allows for inhibition of delay with long but not short CS-US intervals, but in combination with SOP's performance rule deduces CR's with "Weber variability."

Journal Article↗
PubMedDate not supplied

Neural correlates of learning in the prefrontal cortex of the monkey: a predictive model.

The principles underlying the organization and operation of the prefrontal cortex have been addressed by neural network modeling. The involvement of the prefrontal cortex in the temporal organization of behavior can be defined by processing units that switch between two stable states of activity (bistable behavior) in response to synaptic inputs. Long-term representation of programs requiring short-term memory can result from activity-dependent modifications of the synaptic transmission controlling the bistable behavior. After learning, the sustained activity of a given neuron represents the selective memorization of a past event, the selective anticipation of a future event, and the predictability of reinforcement. A simulated neural network illustrates the abilities of the model (1) to learn, via a natural step-by-step training protocol, the paradigmatic task (delayed response) used for testing prefrontal neurons in primates, (2) to display the same categories of neuronal activities, and (3) to predict how they change during learning. In agreement with experimental data, two main types of activity contribute to the adaptive properties of the network. The first is transient activity time-locked to events of the task and its profile remains constant during successive training stages. The second is sustained activity that undergoes nonmonotonic changes with changes in reward contingency that occur during the transition between stages.

Animals↗
PubMed2000

The neuroanatomical correlates of route learning impairment.

Recent functional imaging studies of topographical learning point to the participation of a large network of cortical and subcortical regions. Nevertheless, areas which are crucial remain poorly specified due to the absence of group studies of subjects with focal lesions distributed throughout the brain. We assessed the ability of 127 subjects with stable, focal lesions to learn a complex real-life route, a critical aspect of topographical functioning. Results indicated that impairment in route learning was highly associated with damage to medial occipital and posterior parahippocampal cortices in either hemisphere, the right hippocampus, and the right inferotemporal region. Impairment was seen among 86% of the subjects with damage to any these regions, in contrast to impairment among 31% of subjects with lesions in other regions. The importance of medial occipitotemporal cortices bilaterally and right inferotemporal cortex likely reflects the critical role of the ability to quickly and accurately perceive and learn multiple topographical scenes. The importance of the right (and probably left) posterior parahippocampal gyrus and of the right hippocampus likely reflects their critical, distinctive roles forming an integrated representation of the extended topographical environment (i.e., the appearance of places and spatial relationships between specific places), and consolidating that representation into multifaceted contextual knowledge of the environment.

Brain Mapping↗
PubMed2006

Synaptic interactions underlying song-selectivity in the avian nucleus HVC revealed by dual intracellular recordings.

Stimulus-dependent synaptic interactions underlying selective sensory representations in neural circuits specialized for sensory processing and sensorimotor integration remain poorly understood. The songbird telencephalic nucleus HVC is a sensorimotor area essential to learned vocal control with one projection neuron (PN) type (HVC(RA)) innervating a song premotor pathway, another PN (HVC(X)) innervating a basal ganglia pathway essential to vocal plasticity, and interneurons (HVC(Int)). Playback of the bird's own song (BOS), but not other songs, evokes action potential bursts from both PNs, but HVC(RA) and HVC(X) display distinct BOS-evoked subthreshold responses. To characterize synaptic interactions underlying HVC's BOS-selective responses and assess stimulus-evoked changes in functional interactions between HVC neurons, we made simultaneous in vivo intracellular recordings from various HVC neuron pairs in urethan-anesthetized zebra finches. Spike-triggered averaging revealed that all HVC neuron types receive common excitation and that the onset of this excitation occurs during a narrower time window in projection neurons during BOS playback. To distinguish local from extrinsic contributions to HVC subthreshold response patterns, we inactivated the HVC local circuit with GABA or occluded inhibition in single HVC(X) cells. After either treatment, BOS-evoked responses in HVC(X) neurons became purely depolarizing and subthreshold responses of HVC(X) and HVC(RA) cells became remarkably similar to one another while retaining BOS selectivity. Therefore both PN types receive a common extrinsic source of BOS-selective excitation, and local inhibition specifically alters processing of auditory information in HVC(X) cells. In HVC, excitatory and inhibitory synaptic interactions are recruited in a stimulus-dependent fashion, affecting auditory representations of the BOS locally and in other song nuclei important to song learning and production.

Acoustic Stimulation↗
PubMed1997

The predictive and explanatory power of inductive decision trees: a comparison with artificial neural network learning as applied to the noninvasive diagnosis of coronary artery disease.

BACKGROUND: This paper compares two machine learning systems, an inductive decision tree (IDT) and a back-propagation neural network (ANN), in the noninvasive assessment of coronary artery disease given a set of diagnostic input attributes. A collection of 490 patient cases were accumulated from the reference of diagnostic stress myocardial scintigraphy performed in a nuclear medicine department. All cases had correlating angiography, the results of which were used to derive the target diagnoses. Input attributes included 4 baseline clinical characteristics, 4 nonimaging stress components, and 3 scintigraphic findings. METHODS: We chose 4 possible angiographic criteria for coronary artery disease and assessed the ability of each learning system to develop a diagnostic model. The 2 machine learning systems were compared on the basis of predictive performance and explanatory power. RESULTS: Cross-validation experiments showed the 2 machine learning systems to have equivalent predictive power at the same level as the clinical scan reading. For the 70% stenosis criterion, the IDT had a sensitivity of 94 +/- 3% (mean +/- 95% confidence interval) and a specificity of 59 +/- 8%, and the ANN had a sensitivity of 97 +/- 2% and a specificity of 51 +/- 13%. However the IDT system exhibited excellent explanatory power; producing simple representations of the diagnostic models which agree with previous research. CONCLUSION: In comparison with the more widely used ANNs, the IDT learning system may bring advantages to certain problems in diagnostic classification.

Coronary Angiography↗
PubMed1998

Similarity and the development of rules.

Similarity-based and rule-based accounts of cognition are often portrayed as opposing accounts. In this paper we suggest that in learning and development, the process of comparison can act as a bridge between similarity-based and rule-based processing. We suggest that comparison involves a process of structural alignment and mapping between two representations. This kind of structure-sensitive comparison process--which may be triggered either by experiential or symbolic juxtapositions--has a twofold significance for cognitive development. First, as a learning mechanism, comparison facilitates the grasp of structural commonalities and the abstraction of rules; and, second, as a mechanism for the application and extension of previously acquired knowledge, comparison processes facilitate the application of abstract knowledge to new instances.

Adult↗
PubMed1993

Proprioceptive guidance and motor planning of reaching movements to unseen targets.

The ability to make accurate reaching movements toward proprioceptively defined target locations was studied in seven normal subjects who were trained to reach to five different targets in a horizontal plane, with no vision of hand or target. The task consisted of moving a handle from a fixed origin to each target location, fast and accurately. Target locations were learned in training sessions that utilized acoustic cuing. Most movements were rapid, with a bell-shaped velocity profile. The error in target reproduction, which constituted the difference between the position consciously identified as the correct target location and the real target location, was calculated in each trial. This was compared with the error in preprogrammed reaching, which constituted the difference between the point in space where the initial fast movement toward the target ended and the target location. The absence of significant differences between these two error types indicated that the transformation from an internal representation of target location into a motor program for reaching to it did not introduce an additional reaching error. Learning of target locations was done only with the right hand, yet, reaching of both hands was tested. This allowed a comparison between the subjects' ability to utilize a transformed spatial code (reaching with the untrained hand) and their ability to use a direct sensory-motor code (reaching with the trained hand). While transformation of the spatial code was found to reduce it's accuracy, utilization of this code in motor programming again did not appear to introduce an additional error.

Adult↗