Hideyuki Câteau

Bidirectional plasticity in fast-spiking GABA circuits by visual experience.

Experience-dependent plasticity in the brain requires balanced excitation–inhibition. How individual circuit elements contribute to plasticity outcome in complex neocortical networks remains unknown. Here we report an intracellular analysis of ocular dominance plasticity—the loss of acuity and cortical responsiveness for an eye deprived of vision in early life. Unlike the typical progressive loss of pyramidal-cell bias, direct recording from fast-spiking cells in vivo reveals a counterintuitive initial shift towards the occluded eye followed by a late preference for the open eye, consistent with a spike-timing-dependent plasticity rule for these inhibitory neurons. Intracellular pharmacology confirms a dynamic switch of GABA (γ-aminobutyric acid) impact to pyramidal cells following deprivation in juvenile mice only. Together these results suggest that the bidirectional recruitment of an initially binocular GABA circuit may contribute to experience-dependent plasticity in the developing visual cortex.

A stochastic method to predict the consequence of arbitrary forms of spike-timing-dependent plasticity

Synapses in various neural preparations exhibit spike-timing-dependent plasticity (STDP) with a variety of learning window functions. The window functions determine the magnitude and the polarity of synaptic change according to the time difference of pre- and postsynaptic spikes. Numerical experiments revealed that STDP learning with a single-expo nential window function resulted in a bimodal distribution of synaptic conductances as a consequence of competition between synapses. A slightly modified window function, however, resulted in a unimodal distribution rather than a bimodal distribution. Since various window functions have been observed in neural preparations, we develop a rigorous mathematical method to calculate the conductance distribution for any given window function. Our method is based on the Fokker-Planck equation to determine the conductance distribution and on the Ornstein-Uhlenbeck process to characterize the membrane potential fluctuations. Demonstrating that our method reproduces the known quantitative results of STDP learning, we apply the method to the type of STDP learning found recently in the CA1 region of the rat hippocampus. We find that this learning can result in nearly optimized competition between synapses. Meanwhile, we find that the type of STDP learning found in the cerebellum-like structure of electric fish can result in all-or-none synapses: either all the synaptic conductances are maximized, or none of them becomes significantly large. Our method also determines the window function that optimizes synaptic competition.

Self-organization of memory activity through spike-timing-dependent plasticity.

We studied the self-organization of memory-related activity through spike-timing-dependent plasticity (STDP). Relatively short time windows (∼10 ms) for the plasticity rule give rise to asynchronous persistent activity of low rates (20–30 Hz), which is typically observed in delay periods of working memory task. We demonstrate some network level effects on the activity regulation that cannot be addressed in single-neuron studies. For longer time windows (∼20 ms), the layered cell assemblies that propagate synchronized spikes (synfire chain) are self-organized. Synchronous spike propagation was suggested to underlie the precisely timed spikes in the monkey prefrontal cortex. The present results suggest that the two networks for sustained activity are different realizations of the same principle for synaptic wiring.

Sustained activity with low firing rate in a recurrent network regulated by spike-timing-dependent plasticity

Abstract In order to study roles of spike-timing-dependent plasticity (STDP) at the network level, we applied STDP to a model of the cortical recurrent network. We found that STDP brought self-organization of a bistable neural activity that is essential for the working memory function. Furthermore, our simulations showed that the typical two firing patterns during the persistent activity were achieved, which depend on the time window of STDP; the short time window ( ∼10 ms ) yielded an asynchronous firing activity, whereas the longer one ( ⩾20 ms ) produced a synchronous spike packet propagating along a chain of cell assemblies.

Near scale-free dynamics in neural population activity of waking/sleeping rats revealed by multiscale analysis.

A neuron embedded in an intact brain, unlike an isolated neuron, participates in network activity at various spatial resolutions. Such multiple scale spatial dynamics is potentially reflected in multiple time scales of temporal dynamics. We identify such multiple dynamical time scales of the inter-spike interval (ISI) fluctuations of neurons of waking/sleeping rats by means of multiscale analysis. The time scale of large non-Gaussianity in the ISI fluctuations, measured with the Castaing method, ranges up to several minutes, markedly escaping the low-pass filtering characteristics of neurons. A comparison between neural activity during waking and sleeping reveals that non-Gaussianity is stronger during waking than sleeping throughout the entire range of scales observed. We find a remarkable property of near scale independence of the magnitude correlations as the primary cause of persistent non-Gaussianity. Such scale-invariance of correlations is characteristic of multiplicative cascade processes and raises the possibility of the existence of a scale independent memory preserving mechanism.

An accurate and widely applicable method to determine the distribution of synaptic strengths formed by the spike-timing-dependent learning

Abstract We provide a mathematical method to determine the distribution of synaptic strengths formed by any types of spike-timing-dependent plasticity (STDP). This becomes possible by applying the theory of Ornstein–Uhlenbeck process in determining the Fokker–Planck equation that characterizes the distribution. We verify our novel method by reproducing quantitative properties of STDP observed in previous simulation results. We apply our method to CA1-type window function and electric fish-type window function to demonstrate possible implications of STDP. Moreover, we derive basic properties of STDP from our formalism. Especially, we determine the optimal window function for synaptic competition.

Dendritic Slow Dynamics Enables Localized Cortical Activity to Switch between Mobile and Immobile Modes with Noisy Background Input

Mounting lines of evidence suggest the significant computational ability of a single neuron empowered by active dendritic dynamics. This motivates us to study what functionality can be acquired by a network of such neurons. The present paper studies how such rich single-neuron dendritic dynamics affects the network dynamics, a question which has scarcely been specifically studied to date. We simulate neurons with active dendrites networked locally like cortical pyramidal neurons, and find that naturally arising localized activity – called a bump – can be in two distinct modes, mobile or immobile. The mode can be switched back and forth by transient input to the cortical network. Interestingly, this functionality arises only if each neuron is equipped with the observed slow dendritic dynamics and with in vivo-like noisy background input. If the bump activity is considered to indicate a point of attention in the sensory areas or to indicate a representation of memory in the storage areas of the cortex, this would imply that the flexible mode switching would be of great potential use for the brain as an information processing device. We derive these conclusions using a natural extension of the conventional field model, which is defined by combining two distinct fields, one representing the somatic population and the other representing the dendritic population. With this tool, we analyze the spatial distribution of the degree of after-spike adaptation and explain how we can understand the presence of the two distinct modes and switching between the modes. We also discuss the possible functional impact of this mode-switching ability.

Searching for visual features that explain response variance of face neurons in inferior temporal cortex

Despite a large body of research on response properties of neurons in the inferior temporal (IT) cortex, studies to date have not yet produced quantitative feature descriptions that can predict responses to arbitrary objects. This deficit in the research prevents a thorough understanding of object representation in the IT cortex. Here we propose a fragment-based approach for finding quantitative feature descriptions of face neurons in the IT cortex. The development of the proposed method was driven by the assumption that it is possible to recover features from a set of natural image fragments if the set is sufficiently large. To find the feature from the set, we compared object responses predicted from each fragment and responses of neurons to these objects, and search for the fragment that revealed the highest correlation with neural object responses. Prediction of object responses of each fragment was made by normalizing Euclidian distance between the fragment and each object to 0 to 1 such that the smaller distance gives the higher value. The distance was calculated at the space where images were transformed to a local orientation space by a Gabor filter and a local max operation. The method allowed us to find features with a correlation coefficient between predicted and neural responses of 0.68 on average (number of object stimuli, 104) from among 560,000 feature candidates, reliably explaining differential responses among faces as well as a general preference for faces over to non-face objects. Furthermore, predicted responses of the resulting features to novel object images were significantly correlated with neural responses to these images. Identification of features comprising specific, moderately complex combinations of local orientations and colors enabled us to predict responses to upright and inverted faces, which provided a possible mechanism of face inversion effects. (292/300).

A Method to Construct Visual Recognition Algorithms on the Basis of Neural Activity Data

Visual recognition by animals significantly outperforms man-made algorithms. The brain’s intelligent choice of visual features is considered to be underlying this performance gap. In order to attain better performance for man-made algorithms, we suggest using the visual features that are used in the brain in these algorithms. For this goal, we propose to obtain visual features correlated with the brain activity by applying a kernel canonical correlation analysis (KCCA) method to pairs of image data and neural data recorded from the brain of an animal exposed to the images. It is expected that only the visual features that are highly correlated with the neural activity provide useful information for visual recognition. Applied to hand-written digits as image data and activity data of a multi-layer neural network model as a model for a brain, the method successfully extracted visual features used in the neural network model. Indeed, the use of these visual features in the support vector machine (SVM) made it possible to discriminate the hand-written digits. Since this discrimination required to utilize the knowledge possessed in the neural network model, a simple application of the usual SVM without the use of these features could not discriminate them. We further demonstrate that even the use of non-digit hand-written characters for the KCCA extracts visual features which enable the SVM to discriminate the hand-written digits. This indicates the versatile applicability of our method.

Visual discrimination task with head fixation suggests ability of amodal completion in rats

To investigate the role of uncertainty in action learning, we analyzed rats’ choice behaviors using a Bayesian Q-learning model which considers not only the expectation but also the probability distribution of reward for action choice. In the free choice task, a rat selected either the left or right hole and received a reward of a food pellet stochastically. The reward probabilities of the left and right holes were chosen from 6 settings (high: 100% vs 66%; mid: 66% vs 33%; low: 33% vs 0% for left vs right holes, and the opposites). The rewardprobability setting was changed after the animal chose the more rewarding hole 80% or more in the last 20 trials. Two rats performed 45 sessions, each of which consisted of 290–1210 trials. We analyzed the rats’ choice behaviors using four variants of standard Qlearning, in which the expectation of reward for each action was updated as the action value and used for action choice, and three variants of Bayesian Q-learning, in which a weighted sum of the mean and the standard deviation of the reward distribution was used as the effective action value. Rats’ choice behaviors were significantly better fit by a Bayesian Q-learning model than by standard Q-learning models, suggesting that rats’ choice learning depended of reward uncertainties. The coefficients for the reward standard deviation of the Bayesian Q-learning model were positive in most sessions, suggesting uncertainty seeking. However, further analyses showed that the choice behaviors matched better to the Bayesian Q-learning model only in lowand mid-reward-probability conditions, while the behaviors in high-reward-probability conditions fit better to a standard Q-learning model. These results suggest that rats utilize the uncertainty for their choices depending on the reward condition and that they are uncertainty seeking under lower reward probability conditions. Research fund: PRESTO-JST.