Modeling perceptual learning with multiple interacting elements: A neural network model describing early visual perceptual learning

We introduce a neural network model of an early visual cortical area, in order to understand better results of psychophysical experiments concerning perceptual learning during odd element (pop-out) detection tasks (Ahissar and Hochstein, 1993, 1994a).The model describes a network, composed of orientation selective units, arranged in a hypercolumn structure, with receptive field properties modeled from real monkey neurons. Odd element detection is a final pattern of activity with one (or a few) salient units active. The learning algorithm used was the Associative reward-penalty (Ar-p) algorithm of reinforcement learning (Barto and Anandan, 1985), following physiological data indicating the role of supervision in cortical plasticity.Simulations show that network performance improves dramatically as the weights of inter-unit connections reach a balance between lateral iso-orientation inhibition, and facilitation from neighboring neurons with different preferred orientations. The network is able to learn even from chance performance, and in the presence of a large amount of noise in the response function. As additional tests of the model, we conducted experiments with human subjects in order to examine learning strategy and test model predictions.     

Feedforward architectures driven by inhibitory interactions

Directed information transmission is paramount for many social, physical, and biological systems. For neural systems, scientists have studied this problem under the paradigm of feedforward networks for decades. In most models of feedforward networks, activity is exclusively driven by excitatory neurons and the wiring patterns between them, while inhibitory neurons play only a stabilizing role for the network dynamics. Motivated by recent experimental discoveries of hippocampal circuitry, cortical circuitry, and the diversity of inhibitory neurons throughout the brain, here we illustrate that one can construct such networks even if the connectivity between the excitatory units in the system remains random. This is achieved by endowing inhibitory nodes with a more active role in the network. Our findings demonstrate that apparent feedforward activity can be caused by a much broader network-architectural basis than often assumed.     

Oscillatory interactions between sensorimotor cortex and the periphery

Field potential recordings from motor cortex show oscillations in the beta-band (approximately 20 Hz), which are coherent with similar oscillations in the activity of contralateral contracting muscles. Recent findings have revised concepts of how this activity might be generated in the cortex, suggesting it could achieve useful computation. Other evidence shows that these oscillations engage not just motor structures, but also return from muscle to the central nervous system via feedback afferent pathways. Somatosensory cortex has strong beta-band oscillations, which are synchronised with those in motor cortex, allowing oscillatory sensory reafference to be interpreted in the context of the oscillatory motor command which produced it.     

Lactosylceramide: lateral interactions with cholesterol

Lactosylceramide (LacCer) is a key intermediate in glycosphingolipid metabolism and is highly enriched in detergent-resistant biomembrane fractions associated with microdomains, i.e., rafts and caveolae. Here, the lateral interactions of cholesterol with LacCers containing various homogeneous saturated (8:0, 16:0, 18:0, 24:0) or monounsaturated acyl chains (18:1, 24:1) have been characterized using a Langmuir-type film balance. Cholesterol-induced changes in lateral packing were assessed by measuring changes in average molecular area, i.e., area condensations, and in lateral elasticity, i.e., surface compressional moduli (C S(-1)) with emphasis on high surface pressures (> or = 30 mN/m) that mimic biomembrane conditions. Cholesterol most dramatically affected the lateral packing elasticity of LacCers with long saturated acyl chains at sterol mole fractions > or = 0.3, consistent with liquid-ordered (LO) phase formation. The lateral elasticity within the LacCer-cholesterol LO-phase was much lower than that observed within pure LacCer condensed, i.e., gel, phase. The magnitude of the cholesterol-induced reduction in lateral elasticity was strongly mitigated by cis monounsaturation in the LacCer acyl chain. At identical high sterol mole fractions, higher lateral elasticity was observed within LacCer-cholesterol mixtures compared with galactosylceramide-cholesterol and sphingomyelin-cholesterol mixtures. The results show how changes to sphingolipid headgroup and acyl chain structure contribute to the modulation of lateral packing elasticity in sphingolipid-cholesterol LO-phases.     

Stabilization of Hebbian neural nets by inhibitory learning

In Hebbian neural models synaptic reinforcement occurs when the pre- and post-synaptic neurons are simultaneously active. This causes an instability toward unlimited growth of excitatory synapses. The system can be stabilized by recurrent inhibition via modifiable inhibitory synapses. When this process is included, it is possible to dispense with the non-linear normalization or cut-off conditions which were necessary for stability in previous models. The present formulation is response-linear if synaptic changes are slow. It is self-consistent because the stabilizing effects will tend to keep most neural activity in the middle range, where neural response is approximately linear. The linearized equations are tensor invariant under a class of rotations of the state space. Using this, the response to stimulation may be derived as a set of independent modes of activity distributed over the net, which may be identified with cell assemblies. A continuously infinite set of equivalent solutions exists.     

The N2pc is increased by perceptual learning but is unnecessary for the transfer of learning

Background

Practice improves human performance in many psychophysical paradigms. This kind of improvement is thought to be the evidence of human brain plasticity. However, the changes that occur in the brain are not fully understood.

Methodology/Principal Findings

The N2pc component has previously been associated with visuo-spatial attention. In this study, we used event-related potentials (ERPs) to investigate whether the N2pc component changed during long-term visual perceptual learning. Thirteen subjects completed several days of training in an orientation discrimination task, and were given a final test 30 days later. The results showed that behavioral thresholds significantly decreased across training sessions, and this decrement was also present in the untrained visual field. ERPs showed training significantly increased the N2pc amplitude, and this effect could be maintained for up to 30 days. However, the increase in N2pc was specific to the trained visual field.

Conclusion/Significance

Training caused spatial attention to be increasingly focused on the target positions. However, this process was not transferrable from the trained to the untrained visual field, which suggests that the increase in N2pc may be unnecessary for behavioral improvements in the untrained visual field.     

High-order behaviour in learning gate networks with lateral inhibition

In this work we present a neural network model incorporating activity-dependent presynaptic facilitation with multidimensional inputs. The processing unit used is based on a slightly simplified version of the Learning Gate Model proposed by Ciaccia et al. (1992). The network topology integrates a well-known biological neural circuit with a lateral inhibition connection subnet. By means of simulation experiments, we show that the proposed networks exhibit basic and high-order features of associative learning. In particular, overshadowing and blocking are reproduced in the presence of both noise-free and noisy inputs. The role of noise in the development of high-order learning capabilities is also discussed.This article was processed by the author using the LAT_EX style file pljour2 from Springer-Verlag.     

Auditory cortical contrast enhancing by global winner-take-all inhibitory interactions

Brains decompose the world into discrete objects of perception, thereby facing the problem of how to segregate and selectively address similar objects that are concurrently present in a scene. Theoretical models propose that this could be achieved by neuronal implementations of so-called winner-take-all algorithms where neuronal representations of objects or object features interact in a competitive manner. Here we present evidence for the existence of such a mechanism in an animal species. We present electrophysiological, neuropharmacological and neuroanatomical data which suggest a novel view of the role of GABA(A)-mediated inhibition in primary auditory cortex (AI), where intracortical GABA(A)-mediated inhibition operates on a global scale within a circular map of sound periodicity representation in AI, with functionally inhibitory projections of similar effect from any location throughout the whole map. These interactions could underlie the proposed competitive "winner-take-all" algorithm to support object segregation, e.g., segregation of different speakers in cocktail-party situations.     

Effectively reducing sensory eye dominance with a push-pull perceptual learning protocol

Much knowledge of sensory cortical plasticity is gleaned from perceptual learning studies that improve visual performance [1-7]. Although the improvements are likely caused by modifications of excitatory and inhibitory neural networks, most studies were not primarily designed to differentiate their relative contributions. Here we designed a novel push-pull training protocol to reduce sensory eye dominance (SED), a condition that is mainly caused by unbalanced interocular inhibition [8-10]. During the training, an attention cue presented to the weak eye precedes the binocular competitive stimulation. The cue stimulates the weak eye (push) while causing interocular inhibition of the strong eye (pull). We found that this push-pull protocol reduces SED (shifts the balance toward the weak eye) and improves stereopsis more so than the push-only protocol, which solely stimulates the weak eye without inhibiting the strong eye. The stronger learning effect with the push-pull training than the push-only training underscores the crucial involvement of a putative inhibitory mechanism in sensory plasticity. The design principle of the push-pull protocol can potentially lend itself as an effective, noninvasive treatment of amblyopia.     

Blocked and test-stimulus exposure effects in perceptual learning re-examined

Three experiments re-examined the effects of blocked or alternated exposure to the conditioning and test stimuli and the effect of simple exposure to the test stimulus, on stimulus generalization. In all experiments rats received conditioning where a compound flavor, AX, was paired with LiCl-induced illness. All rats were tested for generalization with another flavor, BX. In Experiment 1, rats that received alternating exposure to the two flavor compounds, AX and BX, prior to conditioning showed less generalization to BX than rats that received no exposure. Exposure to BX or AX alone was also somewhat effective in reducing generalization. In Experiment 2 blocked exposure to AX and BX prior to conditioning was effective in reducing generalization, as was alternated exposure, and extended exposure to BX was more effective than the other procedures. In Experiment 3, exposure to X alone prior to conditioning produced generalization equal to that produced by alternated or blocked exposure and replicated the effect of extended exposure to BX found in the previous experiment. The relevance of the results to the theories proposed by McLaren and Macintosh [Anim. Learn. Behav. 28 (2000) 211] and Hall [Q. J. Exp. Psychol. B 56 (2003) 43] is discussed.     

Metabolic buffering exerted by macromolecular crowding on DNA-DNA interactions: origin and physiological significance

Crowding, which characterizes the interior of all living cells, has been shown to dramatically affect biochemical processes, leading to stabilization of compact morphologies, enhanced macromolecular associations, and altered reaction rates. Due to the crowding-mediated shift in binding equilibria toward association, crowding agents were proposed to act as a metabolic buffer, significantly extending the range of intracellular conditions under which interactions occur. Crowding may, however, impose a liability because, by greatly and generally enhancing macromolecular association, it can lead to irreversible interactions. To better understand the physical determinants and physiological consequences of crowding-mediated buffering, we studied the effects of crowding, or excluded volume, on DNA structures. Results obtained from isothermal titration calorimetry (ITC) and UV melting experiments indicate that crowding-induced effects are marginal under conditions that a priori favor association of DNA strands but become progressively larger when conditions deteriorate. As such, crowding exerts "genuine" buffering activity. Unexpectedly, crowding-mediated effects are found to include enthalpy terms that favorably contribute to association processes. We propose that these enthalpy terms and preferential stabilization derive from a reconfiguration of DNA hydration that occurs in dense DNA-rich phases obtained in crowded environments.     

Dissociation between the Espinet and perceptual learning effects in flavour aversion conditioning

In three experiments rats were given pre-exposure to two compound flavours, AX and BX, the two compounds being presented for some subjects on alternate trials (the intermixed schedule) and, for others, in separate blocks of trials (the blocked schedule). After aversion conditioning with A (in Experiments 1 and 2), the inhibitory properties of B were tested using both retardation (Experiment 1) and summation tests (Experiment 2). The results failed to support the proposal [Anim. Learn. Behav. 23 (1995) 361] that B should acquire inhibitory properties in the intermixed condition (the "Espinet effect"). Experiment 3 demonstrated that generalisation to BX after conditioning with AX was attenuated by intermixed pre-exposure (a perceptual learning effect). This pattern of results challenges the hypothesis that inhibitory learning during intermixed pre-exposure to AX and BX can account for both the Espinet and the perceptual learning effects.     

Retrograde interference in perceptual learning of a peripheral hyperacuity task

Consolidation, a process that stabilizes memory trace after initial acquisition, has been studied for over a century. A number of studies have shown that a skill or memory must be consolidated after acquisition so that it becomes resistant to interference from new information. Previous research found that training on a peripheral 3-dot hyperacuity task could retrogradely interfere with earlier training on the same task but with a mirrored stimulus configuration. However, a recent study failed to replicate this finding. Here we address the controversy by replicating both patterns of results, however, under different experimental settings. We find that retrograde interference occurs when eye-movements are tightly controlled, using a gaze-contingent display, where the peripheral stimuli were only presented when subjects maintained fixation. On the other hand, no retrograde interference was found in a group of subjects who performed the task without this fixation control. Our results provide a plausible explanation of why divergent results were found for retrograde interference in perceptual learning on the 3-dot hyperacuity task and confirm that retrograde interference can occur in this type of low-level perceptual learning. Furthermore, our results demonstrate the importance of eye-movement controls in studies of perceptual learning in the peripheral visual field.     

Neural plasticity in high-level visual cortex underlying object perceptual learning

With intensive training, human can achieve impressive behavioral improvement on various perceptual tasks. This phenomenon, termed perceptual learning, has long been considered as a hallmark of the plasticity of sensory neural system. Not surprisingly, high-level vision, such as object perception, can also be improved by perceptual learning. Here we review recent psychophysical, electrophysiological, and neuroimaging studies investigating the effects of training on object selective cortex, such as monkey inferior temporal cortex and human lateral occipital area. Evidences show that learning leads to an increase in object selectivity at the single neuron level and/or the neuronal population level. These findings indicate that high-level visual cortex in humans is highly plastic and visual experience can strongly shape neural functions of these areas. At the end of the review, we discuss several important future directions in this area.