Synaptic learning rules and sparse coding in a model sensory system

Neural circuits exploit numerous strategies for encoding information. Although the functional significance of individual coding mechanisms has been investigated, ways in which multiple mechanisms interact and integrate are not well understood. The locust olfactory system, in which dense, transiently synchronized spike trains across ensembles of antenna lobe (AL) neurons are transformed into a sparse representation in the mushroom body (MB; a region associated with memory), provides a well-studied preparation for investigating the interaction of multiple coding mechanisms. Recordings made in vivo from the insect MB demonstrated highly specific responses to odors in Kenyon cells (KCs). Typically, only a few KCs from the recorded population of neurons responded reliably when a specific odor was presented. Different odors induced responses in different KCs. Here, we explored with a biologically plausible model the possibility that a form of plasticity may control and tune synaptic weights of inputs to the mushroom body to ensure the specificity of KCs' responses to familiar or meaningful odors. We found that plasticity at the synapses between the AL and the MB efficiently regulated the delicate tuning necessary to selectively filter the intense AL oscillatory output and condense it to a sparse representation in the MB. Activity-dependent plasticity drove the observed specificity, reliability, and expected persistence of odor representations, suggesting a role for plasticity in information processing and making a testable prediction about synaptic plasticity at AL-MB synapses.     

Millisecond-timescale local network coding in the rat primary somatosensory cortex

Correlation among neocortical neurons is thought to play an indispensable role in mediating sensory processing of external stimuli. The role of temporal precision in this correlation has been hypothesized to enhance information flow along sensory pathways. Its role in mediating the integration of information at the output of these pathways, however, remains poorly understood. Here, we examined spike timing correlation between simultaneously recorded layer V neurons within and across columns of the primary somatosensory cortex of anesthetized rats during unilateral whisker stimulation. We used bayesian statistics and information theory to quantify the causal influence between the recorded cells with millisecond precision. For each stimulated whisker, we inferred stable, whisker-specific, dynamic bayesian networks over many repeated trials, with network similarity of 83.3±6% within whisker, compared to only 50.3±18% across whiskers. These networks further provided information about whisker identity that was approximately 6 times higher than what was provided by the latency to first spike and 13 times higher than what was provided by the spike count of individual neurons examined separately. Furthermore, prediction of individual neurons' precise firing conditioned on knowledge of putative pre-synaptic cell firing was 3 times higher than predictions conditioned on stimulus onset alone. Taken together, these results suggest the presence of a temporally precise network coding mechanism that integrates information across neighboring columns within layer V about vibrissa position and whisking kinetics to mediate whisker movement by motor areas innervated by layer V.     

Associative learning in a network model of Hermissenda crassicornis

A time-varying Resistance-Capacitance (RC) circuit computer model was constructed based on known membrane and synaptic properties of the visualvestibular network of the marine snail Hermissenda crassicornis. Specific biophysical properties and synaptic connections of identified neurons are represented as lumped parameters (circuit elements) in the model; in the computer simulation, differential equations are approximated by difference equations. The model's output, membrane potential, an indirect measure of firing frequency, closely parallels the behavioral and electrophysiologic outputs of Hermissenda in response to the same input stimuli presented during and after associative learning. The parallelism of the computer modeled and the biologic outputs suggests that the model captures the features necessary and sufficient for associative learning.     

Efficient sparse coding in early sensory processing: lessons from signal recovery

Sensory representations are not only sparse, but often overcomplete: coding units significantly outnumber the input units. For models of neural coding this overcompleteness poses a computational challenge for shaping the signal processing channels as well as for using the large and sparse representations in an efficient way. We argue that higher level overcompleteness becomes computationally tractable by imposing sparsity on synaptic activity and we also show that such structural sparsity can be facilitated by statistics based decomposition of the stimuli into typical and atypical parts prior to sparse coding. Typical parts represent large-scale correlations, thus they can be significantly compressed. Atypical parts, on the other hand, represent local features and are the subjects of actual sparse coding. When applied on natural images, our decomposition based sparse coding model can efficiently form overcomplete codes and both center-surround and oriented filters are obtained similar to those observed in the retina and the primary visual cortex, respectively. Therefore we hypothesize that the proposed computational architecture can be seen as a coherent functional model of the first stages of sensory coding in early vision.     

Associative learning in a network model of Hermissenda crassicornis

A companion paper in a previous issue of this journal presented a resistance-capacitance circuit computer model of the four-neuron visual-vestibular network of the invertebrate marine mollusk Hermissenda crassicornis. In the present paper, we demonstrate that changes in the model's output in response to simulated associative training is quantitatively similar to behavioral and electrophysiological changes in response to associative training of Hermissenda crassicornis. Specifically, the model demonstrates many characteristics of conditioning: sensitivity to stimulus contingency, stimulus specificity, extinction, and savings. The model's learning features also are shown to be devoid of non-associative components. Thus, this computational model is an excellent tool for examining the information flow and dynamics of biological associative learning and for uncovering insights concerning associative learning, memory, and recall that can be applied to the development of artificial neural networks.     

Associative learning in a neuromimetic network with local competitions

Presented here is a neuromimetic model for the learning of associations between activity patterns originating from recoding layers. These layers are described as networks of cellular clusters made up of competitive formal neurons. A rule of synaptic plasticity with improved neurobiological realism is proposed; it allows for fast learning of large sets of associations.     

Fast coding of orientation in primary visual cortex

Understanding how populations of neurons encode sensory information is a major goal of systems neuroscience. Attempts to answer this question have focused on responses measured over several hundred milliseconds, a duration much longer than that frequently used by animals to make decisions about the environment. How reliably sensory information is encoded on briefer time scales, and how best to extract this information, is unknown. Although it has been proposed that neuronal response latency provides a major cue for fast decisions in the visual system, this hypothesis has not been tested systematically and in a quantitative manner. Here we use a simple 'race to threshold' readout mechanism to quantify the information content of spike time latency of primary visual (V1) cortical cells to stimulus orientation. We find that many V1 cells show pronounced tuning of their spike latency to stimulus orientation and that almost as much information can be extracted from spike latencies as from firing rates measured over much longer durations. To extract this information, stimulus onset must be estimated accurately. We show that the responses of cells with weak tuning of spike latency can provide a reliable onset detector. We find that spike latency information can be pooled from a large neuronal population, provided that the decision threshold is scaled linearly with the population size, yielding a processing time of the order of a few tens of milliseconds. Our results provide a novel mechanism for extracting information from neuronal populations over the very brief time scales in which behavioral judgments must sometimes be made.     

Modeling neuronal dynamic coding in primary visual cortex

The visual system can be considered as a multi-layered and dynamic image processing system. According to experimental evidence, the receptive field (RF) organization is characterized by spatio-temporal properties. The modified extended Gabor (MEG) function model was proposed to describe the main spatio-temporal properties of RF at different levels of visual pathway. Based on the MEG model, a three-layered dynamic coding model was constructed for a complex cell. The responses of the complex cell depend on synaptic events from a simple cell assembly within a time window. The membrane potential evolution equation was applied to the analysis of the length of a time window. The simulation results demonstrated that a complex cell plays as a coincidence detector in encoding synaptic events within the time window.     

Associative search network: A reinforcement learning associative memory

An associative memory system is presented which does not require a teacher to provide the desired associations. For each input key it conducts a search for the output pattern which optimizes an external payoff or reinforcement signal. The associative search network (ASN) combines pattern recognition and function optimization capabilities in a simple and effective way. We define the associative search problem, discuss conditions under which the associative search network is capable of solving it, and present results from computer simulations. The synthesis of sensory-motor control surfaces is discussed as an example of the associative search problem.     

Noise enhances subthreshold oscillations in injured primary sensory neurons

Noise can play a constructive role in the detection of weak signals in various kinds of peripheral receptors and neurons. What the mechanism underlying the effect of noise is remains unclear. Here, the perforated patch-clamp technique was used on isolated cells from chronic compression of the dorsal root ganglion (DRG) model. Our data provided new insight indicating that, under conditions without external signals, noise can enhance subthreshold oscillations, which was observed in a certain type of neurons with high-frequency (20-100 Hz) intrinsic resonance from injured DRG neurons. The occurrence of subthreshold oscillation considerably decreased the threshold potential for generating repetitive firing. The above effects of noise can be abolished by blocking the persistent sodium current (I(Na, P)). Utilizing a mathematical neuron model we further simulated the effect of noise on subthreshold oscillation and firing, and also found that noise can enhance the electrical activity through autonomous stochastic resonance. Accordingly, we propose a new concept of the effects of noise on neural intrinsic activity, which suggests that noise may be an important factor for modulating the excitability of neurons and generation of chronic pain signals.     

Phase-of-firing coding of natural visual stimuli in primary visual cortex

We investigated the hypothesis that neurons encode rich naturalistic stimuli in terms of their spike times relative to the phase of ongoing network fluctuations rather than only in terms of their spike count. We recorded local field potentials (LFPs) and multiunit spikes from the primary visual cortex of anaesthetized macaques while binocularly presenting a color movie. We found that both the spike counts and the low-frequency LFP phase were reliably modulated by the movie and thus conveyed information about it. Moreover, movie periods eliciting higher firing rates also elicited a higher reliability of LFP phase across trials. To establish whether the LFP phase at which spikes were emitted conveyed visual information that could not be extracted by spike rates alone, we compared the Shannon information about the movie carried by spike counts to that carried by the phase of firing. We found that at low LFP frequencies, the phase of firing conveyed 54% additional information beyond that conveyed by spike counts. The extra information available in the phase of firing was crucial for the disambiguation between stimuli eliciting high spike rates of similar magnitude. Thus, phase coding may allow primary cortical neurons to represent several effective stimuli in an easily decodable format.     

Membrane potential synchrony in primary visual cortex during sensory stimulation

When the primary visual cortex (V1) is activated by sensory stimulation, what is the temporal correlation between the synaptic inputs to nearby neurons? This question underlies the origin of correlated activity, the mechanism of how visually evoked activity emerges and propagates in cortical circuits, and the relationship between spontaneous and evoked activity. Here, we have recorded membrane potential from pairs of V1 neurons in anesthetized cats and found that visual stimulation suppressed low-frequency membrane potential synchrony (0-10 Hz), and often increased synchrony at high frequencies (20-80 Hz). The increase in high-frequency synchrony occurred for neurons with similar orientation preferences and for neurons with different orientation preferences and occurred for a wide range of stimulus orientations. Thus, while only a subset of neurons spike in response to visual stimulation, a far larger proportion of the circuit is correlated with spiking activity through subthreshold, high-frequency synchronous activity that crosses functional domains.     

Strong single-fiber sensory inputs to olfactory cortex: implications for olfactory coding

Olfactory information is first encoded in a combinatorial fashion by olfactory bulb glomeruli, which individually represent distinct chemical features of odors. This information is then transmitted to piriform (olfactory) cortex, via axons of olfactory bulb mitral and tufted (M/T) cells, where it is presumed to form the odor percept. However, mechanisms governing the integration of sensory information in mammalian olfactory cortex are unclear. Here we show that single M/T cells can make powerful connections with cortical pyramidal cells, and coincident input from few M/T cells is sufficient to elicit spike output. These findings suggest that odor coding is broad and distributed in olfactory cortex.     

Associative learning in biochemical networks

It has been recently suggested that there are likely generic features characterizing the emergence of systems constructed from the self-organization of self-replicating agents acting under one or more selection pressures. Therefore, structures and behaviors at one length scale may be used to infer analogous structures and behaviors at other length scales. Motivated by this suggestion, we seek to characterize various "animate" behaviors in biochemical networks, and the influence that these behaviors have on genomic evolution. Specifically, in this paper, we develop a simple, chemostat-based model illustrating how a process analogous to associative learning can occur in a biochemical network. Associative learning is a form of learning whereby a system "learns" to associate two stimuli with one another. Associative learning, also known as conditioning, is believed to be a powerful learning process at work in the brain (associative learning is essentially "learning by analogy"). In our model, two types of replicating molecules, denoted as A and B, are present in some initial concentration in the chemostat. Molecules A and B are stimulated to replicate by some growth factors, denoted as G(A) and G(B), respectively. It is also assumed that A and B can covalently link, and that the conjugated molecule can be stimulated by either the G(A) or G(B) growth factors (and can be degraded). We show that, if the chemostat is stimulated by both growth factors for a certain time, followed by a time gap during which the chemostat is not stimulated at all, and if the chemostat is then stimulated again by only one of the growth factors, then there will be a transient increase in the number of molecules activated by the other growth factor. Therefore, the chemostat bears the imprint of earlier, simultaneous stimulation with both growth factors, which is indicative of associative learning. It is interesting to note that the dynamics of our model is consistent with certain aspects of Pavlov's original series of conditioning experiments in dogs. We discuss how associative learning can potentially be performed in vitro within RNA, DNA, or peptide networks. We also describe how such a mechanism could be involved in genomic evolution, and suggest relevant bioinformatics studies that could potentially resolve these issues.     

Perceptual learning and dynamic changes in primary visual cortex

Perceptual learning is the improved performance that follows practice in a perceptual task. In this issue of Neuron, Yotsumoto et al. use fMRI to show that stimuli presented at the location used in training initially evoke greater activation in primary visual cortex than stimuli presented elsewhere, but this difference disappears once learning asymptotes.     

The cutaneous rabbit illusion affects human primary sensory cortex somatotopically

We used functional magnetic resonance imaging (fMRI) to study neural correlates of a robust somatosensory illusion that can dissociate tactile perception from physical stimulation. Repeated rapid stimulation at the wrist, then near the elbow, can create the illusion of touches at intervening locations along the arm, as if a rabbit hopped along it. We examined brain activity in humans using fMRI, with improved spatial resolution, during this version of the classic cutaneous rabbit illusion. As compared with control stimulation at the same skin sites (but in a different order that did not induce the illusion), illusory sequences activated contralateral primary somatosensory cortex, at a somatotopic location corresponding to the filled-in illusory perception on the forearm. Moreover, the amplitude of this somatosensory activation was comparable to that for veridical stimulation including the intervening position on the arm. The illusion additionally activated areas of premotor and prefrontal cortex. These results provide direct evidence that illusory somatosensory percepts can affect primary somatosensory cortex in a manner that corresponds somatotopically to the illusory percept.     

Information coding in a laminar computational model of cat primary visual cortex

Neural populations across cortical layers perform different computational tasks. However, it is not known whether information in different layers is encoded using a common neural code or whether it depends on the specific layer. Here we studied the laminar distribution of information in a large-scale computational model of cat primary visual cortex. We analyzed the amount of information about the input stimulus conveyed by the different representations of the cortical responses. In particular, we compared the information encoded in four possible neural codes: (1) the information carried by the firing rate of individual neurons; (2) the information carried by spike patterns within a time window; (3) the rate-and-phase information carried by the firing rate labelled by the phase of the Local Field Potentials (LFP); (4) the pattern-and-phase information carried by the spike patterns tagged with the LFP phase. We found that there is substantially more information in the rate-and-phase code compared with the firing rate alone for low LFP frequency bands (less than 30 Hz). When comparing how information is encoded across layers, we found that the extra information contained in a rate-and-phase code may reach 90 % in Layer 4, while in other layers it reaches only 60 %, compared to the information carried by the firing rate alone. These results suggest that information processing in primary sensory cortices could rely on different coding strategies across different layers.     

High-precision coding in visual cortex

1. Download : Download high-res image (195KB)
2. Download : Download full-size image

     

Population coding in somatosensory cortex

Computational analyses have begun to elucidate which components of somatosensory cortical population activity may encode basic stimulus features. Recent results from rat barrel cortex suggest that the essence of this code is not synergistic spike patterns, but rather the precise timing of single neuron's first post-stimulus spikes. This may form the basis for a fast, robust population code.