Deciphering a neural code for vision

Deciphering the information that eyes, ears, and other sensory organs transmit to the brain is important for understanding the neural basis of behavior. Recordings from single sensory nerve cells have yielded useful insights, but single neurons generally do not mediate behavior; networks of neurons do. Monitoring the activity of all cells in a neural network of a behaving animal, however, is not yet possible. Taking an alternative approach, we used a realistic cell-based model to compute the ensemble of neural activity generated by one sensory organ, the lateral eye of the horseshoe crab, Limulus polyphemus. We studied how the neural network of this eye encodes natural scenes by presenting to the model movies recorded with a video camera mounted above the eye of an animal that was exploring its underwater habitat. Model predictions were confirmed by simultaneously recording responses from single optic nerve fibers of the same animal. We report here that the eye transmits to the brain robust “neural images” of objects having the size, contrast, and motion of potential mates. The neural code for such objects is not found in ambiguous messages of individual optic nerve fibers but rather in patterns of coherent activity that extend over small ensembles of nerve fibers and are bound together by stimulus motion. Integrative properties of neurons in the first synaptic layer of the brain appear well suited to detecting the patterns of coherent activity. Neural coding by this relatively simple eye helps explain how horseshoe crabs find mates and may lead to a better understanding of how more complex sensory organs process information.

Code-specific policy gradient rules for spiking neurons

Although it is widely believed that reinforcement learning is a suitable tool for describing behavioral learning, the mechanisms by which it can be implemented in networks of spiking neurons are not fully understood. Here, we show that different learning rules emerge from a policy gradient approach depending on which features of the spike trains are assumed to influence the reward signals, i.e., depending on which neural code is in effect. We use the framework of Williams (1992) to derive learning rules for arbitrary neural codes. For illustration, we present policy-gradient rules for three different example codes - a spike count code, a spike timing code and the most general "full spike train" code - and test them on simple model problems. In addition to classical synaptic learning, we derive learning rules for intrinsic parameters that control the excitability of the neuron. The spike count learning rule has structural similarities with established Bienenstock-Cooper-Munro rules. If the distribution of the relevant spike train features belongs to the natural exponential family, the learning rules have a characteristic shape that raises interesting prediction problems.

Code-specific learning rules improve action selection by populations of spiking neurons

Population coding is widely regarded as a key mechanism for achieving reliable behavioral decisions. We previously introduced reinforcement learning for population-based decision making by spiking neurons. Here we generalize population reinforcement learning to spike-based plasticity rules that take account of the postsynaptic neural code. We consider spike/no-spike, spike count and spike latency codes. The multi-valued and continuous-valued features in the postsynaptic code allow for a generalization of binary decision making to multi-valued decision making and continuous-valued action selection. We show that code-specific learning rules speed up learning both for the discrete classification and the continuous regression tasks. The suggested learning rules also speed up with increasing population size as opposed to standard reinforcement learning rules. Continuous action selection is further shown to explain realistic learning speeds in the Morris water maze. Finally, we introduce the concept of action perturbation as opposed to the classical weight- or node-perturbation as an exploration mechanism underlying reinforcement learning. Exploration in the action space greatly increases the speed of learning as compared to exploration in the neuron or weight space.

Neural population coding is optimized by discrete tuning curves

The sigmoidal tuning curve that maximizes the mutual information for a Poisson neuron, or population of Poisson neurons, is obtained. The optimal tuning curve is found to have a discrete structure that results in a quantization of the input signal. The number of quantization levels undergoes a hierarchy of phase transitions as the length of the coding window is varied. We postulate, using the mammalian auditory system as an example, that the presence of a subpopulation structure within a neural population is consistent with an optimal neural code.

Perceptual learning of acoustic noise generates memory-evoked potentials

Experience continuously imprints on the brain at all stages of life. The traces it leaves behind can produce perceptual learning [1], which drives adaptive behavior to previously encountered stimuli. Recently, it has been shown that even random noise, a type of sound devoid of acoustic structure, can trigger fast and robust perceptual learning after repeated exposure [2]. Here, by combining psychophysics, electroencephalography (EEG), and modeling, we show that the perceptual learning of noise is associated with evoked potentials, without any salient physical discontinuity or obvious acoustic landmark in the sound. Rather, the potentials appeared whenever a memory trace was observed behaviorally. Such memory-evoked potentials were characterized by early latencies and auditory topographies, consistent with a sensory origin. Furthermore, they were generated even on conditions of diverted attention. The EEG waveforms could be modeled as standard evoked responses to auditory events (N1-P2) [3], triggered by idiosyncratic perceptual features acquired through learning. Thus, we argue that the learning of noise is accompanied by the rapid formation of sharp neural selectivity to arbitrary and complex acoustic patterns, within sensory regions. Such a mechanism bridges the gap between the short-term and longer-term plasticity observed in the learning of noise [2, 4-6]. It could also be key to the processing of natural sounds within auditory cortices [7], suggesting that the neural code for sound source identification will be shaped by experience as well as by acoustics.

Codage de l’information visuelle par la plasticité et la synchronisation des réponses neuronales dans le cortex visuel primaire du chat

Les systèmes sensoriels encodent l’information sur notre environnement sous la forme d’impulsions électriques qui se propagent dans des réseaux de neurones. Élucider le code neuronal – les principes par lesquels l’information est représentée dans l’activité des neurones – est une question fondamentale des neurosciences. Cette thèse constituée de 3 études (E) s’intéresse à deux types de codes, la synchronisation et l’adaptation, dans les neurones du cortex visuel primaire (V1) du chat. Au niveau de V1, les neurones sont sélectifs pour des propriétés comme l’orientation des contours, la direction et la vitesse du mouvement. Chaque neurone ayant une combinaison de propriétés pour laquelle sa réponse est maximale, l’information se retrouve distribuée dans différents neurones situés dans diverses colonnes et aires corticales. Un mécanisme potentiel pour relier l’activité de neurones répondant à des items eux-mêmes reliés (e.g. deux contours appartenant au même objet) est la synchronisation de leur activité. Cependant, le type de relations potentiellement encodées par la synchronisation n’est pas entièrement clair (E1). Une autre stratégie de codage consiste en des changements transitoires des propriétés de réponse des neurones en fonction de l’environnement (adaptation). Cette plasticité est présente chez le chat adulte, les neurones de V1 changeant d’orientation préférée après exposition à une orientation non préférée. Cependant, on ignore si des neurones spatialement proches exhibent une plasticité comparable (E2). Finalement, nous avons étudié la dynamique de la relation entre synchronisation et plasticité des propriétés de réponse (E3). Résultats principaux — (E1) Nous avons montré que deux stimuli en mouvement soit convergent soit divergent élicitent plus de synchronisation entre les neurones de V1 que deux stimuli avec la même direction. La fréquence de décharge n’était en revanche pas différente en fonction du type de stimulus. Dans ce cas, la synchronisation semble coder pour la relation de cocircularité dont le mouvement convergent (centripète) et divergent (centrifuge) sont deux cas particuliers, et ainsi pourrait jouer un rôle dans l’intégration des contours. Cela indique que la synchronisation code pour une information qui n’est pas présente dans la fréquence de décharge des neurones. (E2) Après exposition à une orientation non préférée, les neurones changent d’orientation préférée dans la même direction que leurs voisins dans 75% des cas. Plusieurs propriétés de réponse des neurones de V1 dépendent de leur localisation dans la carte fonctionnelle corticale pour l’orientation. Les comportements plus diversifiés des 25% de neurones restants sont le fait de différences fonctionnelles que nous avons observé et qui suggèrent une localisation corticale particulière, les singularités, tandis que la majorité des neurones semblent situés dans les domaines d’iso-orientation. (E3) Après adaptation, les paires de neurones dont les propriétés de réponse deviennent plus similaires montrent une synchronisation accrue. Après récupération, la synchronisation retourne à son niveau initial. Par conséquent, la synchronisation semble refléter de façon dynamique la similarité des propriétés de réponse des neurones. Conclusions — Cette thèse contribue à notre connaissance des capacités d’adaptation de notre système visuel à un environnement changeant. Nous proposons également des données originales liées au rôle potentiel de la synchronisation. En particulier, la synchronisation semble capable de coder des relations entre objets similaires ou dissimilaires, suggérant l’existence d’assemblées neuronales superposées.

Single Synapse Information Coding in Intraburst Spike Patterns of Central Pattern Generator Motor Neurons

Burst firing is ubiquitous in nervous systems and has been intensively studied in central pattern generators (CPGs). Previous works have described subtle intraburst spike patterns (IBSPs) that, despite being traditionally neglected for their lack of relation to CPG motor function, were shown to be cell-type specific and sensitive to CPG connectivity. Here we address this matter by investigating how a bursting motor neuron expresses information about other neurons in the network. We performed experiments on the crustacean stomatogastric pyloric CPG, both in control conditions and interacting in real-time with computer model neurons. The sensitivity of postsynaptic to presynaptic IBSPs was inferred by computing their average mutual information along each neuron burst. We found that details of input patterns are nonlinearly and inhomogeneously coded through a single synapse into the fine IBSPs structure of the postsynaptic neuron following burst. In this way, motor neurons are able to use different time scales to convey two types of information simultaneously: muscle contraction (related to bursting rhythm) and the behavior of other CPG neurons (at a much shorter timescale by using IBSPs as information carriers). Moreover, the analysis revealed that the coding mechanism described takes part in a previously unsuspected information pathway from a CPG motor neuron to a nerve that projects to sensory brain areas, thus providing evidence of the general physiological role of information coding through IBSPs in the regulation of neuronal firing patterns in remote circuits by the CNS.

Recording from Two Neurons: Second-Order Stimulus Reconstruction from Spike Trains and Population Coding

We study the reconstruction of visual stimuli from spike trains, representing the reconstructed stimulus by a Volterra series up to second order. We illustrate this procedure in a prominent example of spiking neurons, recording simultaneously from the two H1 neurons located in the lobula plate of the fly Chrysomya megacephala. The fly views two types of stimuli, corresponding to rotational and translational displacements. Second-order reconstructions require the manipulation of potentially very large matrices, which obstructs the use of this approach when there are many neurons. We avoid the computation and inversion of these matrices using a convenient set of basis functions to expand our variables in. This requires approximating the spike train four-point functions by combinations of two-point functions similar to relations, which would be true for gaussian stochastic processes. In our test case, this approximation does not reduce the quality of the reconstruction. The overall contribution to stimulus reconstruction of the second-order kernels, measured by the mean squared error, is only about 5% of the first-order contribution. Yet at specific stimulus-dependent instants, the addition of second-order kernels represents up to 100% improvement, but only for rotational stimuli. We present a perturbative scheme to facilitate the application of our method to weakly correlated neurons.

The structure and precision of retinal spike trains

Assessing the reliability of neuronal spike trains is fundamental to an understanding of the neural code. We measured the reproducibility of retinal responses to repeated visual stimuli. In both tiger salamander and rabbit, the retinal ganglion cells responded to random flicker with discrete, brief periods of firing. For any given cell, these firing events covered only a small fraction of the total stimulus time, often less than 5%. Firing events were very reproducible from trial to trial: the timing jitter of individual spikes was as low as 1 msec, and the standard deviation in spike count was often less than 0.5 spikes. Comparing the precision of spike timing to that of the spike count showed that the timing of a firing event conveyed several times more visual information than its spike count. This sparseness and precision were general characteristics of ganglion cell responses, maintained over the broad ensemble of stimulus waveforms produced by random flicker, and over a range of contrasts. Thus, the responses of retinal ganglion cells are not properly described by a firing probability that varies continuously with the stimulus. Instead, these neurons elicit discrete firing events that may be the fundamental coding symbols in retinal spike trains.

Signals and signs in the nervous system: The dynamic anatomy of electrical activity is probably information-rich

The dichotomy between two groups of workers on neuroelectrical activity is retarding progress. To study the interrelations between neuronal unit spike activity and compound field potentials of cell populations is both unfashionable and technically challenging. Neither of the mutual disparagements is justified: that spikes are to higher functions as the alphabet is to Shakespeare and that slow field potentials are irrelevant epiphenomena. Spikes are not the basis of the neural code but of multiple codes that coexist with nonspike codes. Field potentials are mainly information-rich signs of underlying processes, but sometimes they are also signals for neighboring cells, that is, they exert influence. This paper concerns opportunities for new research with many channels of wide-band (spike and slow wave) recording. A wealth of structure in time and three-dimensional space is different at each scale—micro-, meso-, and macroactivity. The depth of our ignorance is emphasized to underline the opportunities for uncovering new principles. We cannot currently estimate the relative importance of spikes and synaptic communication vs. extrasynaptic graded signals. In spite of a preponderance of literature on the former, we must consider the latter as probably important. We are in a primitive stage of looking at the time series of wide-band voltages in the compound, local field, potentials and of choosing descriptors that discriminate appropriately among brain loci, states (functions), stages (ontogeny, senescence), and taxa (evolution). This is not surprising, since the brains in higher species are surely the most complex systems known. They must be the greatest reservoir of new discoveries in nature. The complexity should not deter us, but a dose of humility can stimulate the flow of imaginative juices.

Medida de uma codificação de estímulos sonoros por potenciais pós-sinápticos e potenciais de campo local usando teoria da informação

Coding process is a fundamental aspect of cerebral functioning. The sensory stimuli transformation in neurophysiological responses has been a research theme in several areas of Neuroscience. One of the most used ways to measure a neural code e ciency is by the use of Information Theory measures, such as mutual information. Using these tools, recent studies show that in the auditory cortex both local eld potentials (LFPs) and action potential spiking times code information about sound stimuli. However, there are no studies applying Information Theory tools to investigate the e ciency of codes that use postsynaptics potentials (PSPs), alone and associated with LFP analysis. These signals are related in the sense that LFPs are partly created by joint action of several PSPs. The present dissertation reports information measures between PSP and LFP responses obtained in the primary auditory cortex of anaesthetized rats and auditory stimuli of distinct frequencies. Our results show that PSP responses hold information about sound stimuli in comparable levels and even greater than LFP responses. We have also found that PSPs and LFPs code sound information independently, since the joint analysis of these signals did neither show synergy nor redundancy.

A neural network based tool for semi-automatic code transformation

A neural network based tool has been developed to assist in the process of code transformation. The tool offers advice on appropriate transformations within a knowledge-driven, semi-automatic parallelisation environment. We have identified the essential characteristics of codes relevant to loop transformations. A Kohonen network is used to discover structure in the characterised codes thus revealing new knowledge that may be brought to bear on the mapping between codes and transformations or transformation sequences. A transform selector based on this process has been developed and successfully applied to the parallelisation of sequential codes.

The role of stochasticity in an information-optimal neural population code

In this paper we consider the optimisation of Shannon mutual information (MI) in the context of two model neural systems The first is a stochastic pooling network (population) of McCulloch-Pitts (MP) type neurons (logical threshold units) subject to stochastic forcing; the second is (in a rate coding paradigm) a population of neurons that each displays Poisson statistics (the so called 'Poisson neuron'). The mutual information is optimised as a function of a parameter that characterises the 'noise level'-in the MP array this parameter is the standard deviation of the noise, in the population of Poisson neurons it is the window length used to determine the spike count. In both systems we find that the emergent neural architecture and; hence, code that maximises the MI is strongly influenced by the noise level. Low noise levels leads to a heterogeneous distribution of neural parameters (diversity), whereas, medium to high noise levels result in the clustering of neural parameters into distinct groups that can be interpreted as subpopulations In both cases the number of subpopulations increases with a decrease in noise level. Our results suggest that subpopulations are a generic feature of an information optimal neural population.

Flow Regime Identification Methodology with Mcnp-X Code And Artificial Neural Network

This paper presents flow regimes identification methodology in multiphase system in annular, stratified and homogeneous oil-water-gas regimes. The principle is based on recognition of the pulse height distributions (PHD) from gamma-ray with supervised artificial neural network (ANN) systems. The detection geometry simulation comprises of two NaI(Tl) detectors and a dual-energy gamma-ray source. The measurement of scattered radiation enables the dual modality densitometry (DMD) measurement principle to be explored. Its basic principle is to combine the measurement of scattered and transmitted radiation in order to acquire information about the different flow regimes. The PHDs obtained by the detectors were used as input to ANN. The data sets required for training and testing the ANN were generated by the MCNP-X code from static and ideal theoretical models of multiphase systems. The ANN correctly identified the three different flow regimes for all data set evaluated. The results presented show that PHDs examined by ANN may be applied in the successfully flow regime identification.