Search

Functional connectivity between cerebral cortical regions during natural sleep has attracted a keen interest from both cognitive and clinical neuroscientists because of its importance in understanding the default mode network of human brain. Multiple recordings of functional near-infrared spectroscopy (fNIRS) in several sleep phases make it possible for us to detect potential differences of directional interactions between cortical areas from healthy subjects and patients with ADHD or sleep disorders. Namely we propose a computational method to estimate time-domain Granger causality among fNIRS time series using a Kolmogorov–Smirnov test based on F-statistics. In order to validate indication of directional interactions, we also apply convergent cross-mapping to the time series as an alternative approach to causality based on state space reconstruction of dynamical systems. Comparing the averaged heatmaps of significant causal pairs of regions, we show that the map of directional interactions varies for each sleep phase, e.g., REM, of the same subject. The observation suggests an unexplored source for non-invasive classification benchmark of the above cognitive disorders.

We investigate the effects of network topology on the dynamical activity of a hierarchically organized network of simulated spiking neurons. With a fixed basic two-by-two grid structure of processing modules each composed by almost 6000 leaky integrate-and-fire neurons and different connectivity schemes in between these modules, we study how the activation and the biologically inspired processes of plasticity on the network shape its topology using invariants based on algebro-topological constructions. By definition, a clique is a fully connected directed subnetwork that means there is one source and one sink in the subnetwork. We define ‘k-clique hub cells’ for a positive integer k any cell which is sink and source cell of at least k 3-cliques. We show that there is a statistically different distribution of in- and out-degrees between clique hubs and other cells. Furthermore, we show that by identifying ‘clique hub cells’ we can find synfire chains that are involved in spatio-temporal firing patterns. Hence, the results suggest a link exists between an initial topological structure characterized by subnetworks cliques and a functional connectivity emerging at a later stage as the outcome of synaptic plasticity mechanisms.

We characterize the attractor-based expressive power of several models of recurrent neural networks.The deterministic rational-weighted networks are Muller Turing equivalent.The deterministic real-weighted and evolving networks recognize the class of B C ( ź 2 0 ) neural ω languages.The nondeterministic rational and real networks recognize the class of Σ 1 1 neural ω-languages. We provide a characterization of the expressive powers of several models of deterministic and nondeterministic first-order recurrent neural networks according to their attractor dynamics. The expressive power of neural nets is expressed as the topological complexity of their underlying neural ω-languages, and refers to the ability of the networks to perform more or less complicated classification tasks via the manifestation of specific attractor dynamics. In this context, we prove that most neural models under consideration are strictly more powerful than Muller Turing machines. These results provide new insights into the computational capabilities of recurrent neural networks.

We consider a simplified Boolean model of the basal ganglia-thalamocortical network, and study the effect of a spike-timing-dependent plasticity (STDP) rule on the stabilization of its attractor dynamics. More precisely, we introduce an adaptive STDP rule which constantly updates its learning rate based on the attractors that the network encounters during a window of past time steps. This so-called network memory is assumed to be dynamic: its duration is step-wise increased every time a trigger input pattern is detected, and is decreased otherwise. In this context, we show that well-adjusted trigger inputs can fine tune the network memory and its associated STDP rule in such a way to drive the network into stable and rich attractor dynamics. We discuss how this feature might be related to reward learning processes in the neurobiological context.

We introduce an analysis method for electroencephalography (EEG) data, focused on Event-Related Potentials (ERPs). Our approach is unsupervised and makes use of a fuzzy clustering algorithm based on the possibilistic framework, and includes a data-driven noise and artifact rejection phase. Our contribution provides a general analysis tool, applicable to any ERP data set, which can uncover the data set's internal structure. The fuzzy clustering algorithm is the core of our method, since its fine-grained membership grades how much a sample belongs to a given cluster, making the method applicable even when groups have a certain overlap. Prior to the clustering step, we apply weights to the feature vectors, optimizing them in order to enhance the variance within the dataset, and we extract time-window interval based features inspired by interval arithmetic. We apply the data processing workflow to the analysis a set of ERPs recorded during an emotional Go/NoGo task. We evaluate the performance of the unsupervised analysis by computing a measure based on the clusterization rate of trials in different experimental conditions. The results on the studied data set show that the proposed method obtains a difference of clusterization rate of 69% in Go vs. NoGo trials, when weights and interval-features are applied to the data, improving previous work not including weights and interval-features which had a rate of 31%. Furthermore, when compared with the standard Fuzzy c-means, our proposed possibilistic clustering algorithm outperforms it in terms of clusterization rate. We also examine the effect of pre-processing the data with Independent Component Analysis and removing noise-related components, and observe that this does not improve significantly the obtained results. These findings demonstrate that our proposed method provides a valuable data processing workflow robust to EEG artifacts and able to produce a clustering that is coherent with the experimental conditions represented in the ERP dataset.

Anaerobic digestion is a natural process that transforms organic substrates to methane and other products. Under controlled conditions the process has been widely applied to manage organic wastes. Improvements in process control are expected to lead to improvements in the technical and economic efficiency of the process. This paper presents and compares 3 different neural network model architectures for use as anaerobic digestion process predictive models. The models predict the future biogas production trend from measured physical and chemical parameters. The first model features an LSTM layer, the second model features a 1-D convolutional layer and the third model combines 2 separate inputs and parallel treatment using LSTM and 1-D convolutional layers followed by merging to produce a single prediction. The predictions can be used to adaptively adjust the substrate feeding rate in accordance with the transient state of the digestion process as defined by liquid feeding rate, the organic acid and ammonium ion concentrations and the pH of the digester liquid phase. The training and testing data were obtained during 1 year of continuous operation of a pilot-plant treating restaurant wastes. PLS regression and ICA were used to select the most relevant process parameters from the data. The 1-D Convolutional based model comprising 272 trainable parameters predicted the future biogas flow rate changes with accuracy as high as 89% and an average accuracy of 58% . The work-flow can be applied to optimize the control of the study digester and to control bioreactors in general.

Studies of Boolean recurrent neural networks are briefly introduced with an emphasis on the attractor dynamics determined by the sequence of distinct attractors observed in the limit cycles. We apply this framework to a simplified model of the basal ganglia-thalamocortical circuit where each brain area is represented by a "neuronal" node in a directed graph. Control parameters ranging from neuronal excitability that affects all cells to targeted local connections modified by a new adaptive plasticity rule, and the regulation of the interactive feedback affecting the external input stream of information, allow the network dynamics to switch between stable domains delimited by highly discontinuous boundaries and reach very high levels of complexity with specific configurations. The significance of this approach with regard to brain circuit studies is briefly discussed.

In the analysis of raw electrophysiological recordings, signal filtering is a crucial step to eliminate frequency components associated with noise and recording artifacts. Two problems have to be addressed: determining the optimal frequency range of the signal and which frequency values (extreme values, maximum and minimum) that are characteristic of the raw signal power spectrum. We developed an algorithm called ERFo (extractor of range for filtering optimization) that determines the frequency range boundaries expected to be optimal for the observation of the spectrum with the largest power in the range of interest. The regular differentiations (first and second derivatives) of the raw signal are used to detect the maximum and minimum amplitudes, which are reported in the algorithm to determine the frequency range of the raw signal filtration.

The attractor-based complexity of a Boolean neural network refers to its ability to discriminate among the possible input streams, by means of alternations between meaningful and spurious attractor dynamics. The higher the complexity, the greater the computational power of the network. The fine tuning of the interactivity – the network’s feedback output combined with the current input stream – can maintain a high degree of complexity within stable domains of the parameters’ space. In addition, the attractor-based complexity of the network is related to the degree of discrimination of specific input streams. We present a novel supervised attractor-based learning procedure aimed at achieving a maximal discriminability degree of a selected input stream. With a predefined target value of discriminability degree and in the absence of changes in the internal connectivity matrix of the network, the learning procedure updates solely the weights of the feedback projections. In a Boolean model of the basal ganglia-thalamocortical circuit, we show how the learning trajectories starting from different configurations can converge to final configurations associated with same high discriminability degree. We discuss the possibility that the limbic system may play the role of the interactive feedback to the network studied here.

A biologically-plausible simulation of a neuronal network is studied as its topology is shaped by its activity by means of an encoding of its connectivity structure as a directed clique complex. Specially defined invariants of this mathematical structure, including the information about synaptic strength, are introduced and show how the initial topology of a network and its evolution during the simulation are tightly inter-related with the dynamical activity.

Neuroheuristic Research Group, University of LausanneQuartier Dorigny, CH-1015 Lausanne, Switzerland, Email: avilla@neuroheuristic.orgAbstract —The concept of interdependent communica-tions systems and Wiener's assertion that a machine thatchanges its responses based on feedback is a machine thatlearns, denes the brain as a cybernetic machine. Systemstheory has traditionally focused on the structure of systemsand their models, whereas cybernetics has focused on howsystems function, how they control their actions, how theycommunicate with other systems or with their own compo-nents. However, structure and function of a system cannotbe understood in separation and cybernetics and systemstheoryshouldbeviewedastwofacetsofasingleapproach,dened as the neuroheuristic approach.1. IntroductionNorbert Wiener, a mathematician, engineer and socialphilosopher, coinedtheword”cybernetics”fromtheGreekword meaning ”steersman”. He dened it as the science ofcontrol and communication in the animal and the machine[1]. Many other denitions have followed since then, butingeneralcyberneticstakesasitsdomainthedesignordis-coveryandapplicationofprinciplesofregulationandcom-munication. Early work sought to dene and apply prin-ciples by which systems may be controlled. More recentwork has attempted to understand how systems describethemselves, control themselves, and organize themselves.The cerebral cortex is not a single entity but an impres-sive network formed by an order of tens of millions of neu-rons, most of them excitatory, and by about ten times moreglial cells. Ninety percent of the inputs received by a cor-tical area come from other areas of the cerebral cortex. Asa whole, the cerebral cortex can be viewed as a machinetalking to itself and could be seen as one big feedback sys-temsubjecttotherelentlessadvanceofentropy,whichsub-verts the exchange of messages that is essential to contin-ued existence (Wiener, 1954). This concept of interdepen-dent communications systems, also known as systems the-ory, coupled with Wiener's assertion that a machine thatchanges its responses based on feedback is a machine thatlearns, denes the cerebral cortex as a cybernetic machine.Therefore, the focus of investigation is shifted from com-munication and control to interaction. Systems theory hastraditionally focused more on the structure of systems andtheir models, whereas cybernetics has focused more onhow systems function, that is to say how they control theiractions, how they communicate with other systems or withtheir own components. However, structure and function ofa system cannot be understood in separation and cybernet-ics and systems theory should be viewed as two facets of asingle approach, dened as “the neuroheuristic approach”.2. Classical and Interactive ComputationMcCulloch and Pitts [2] proposed a modelization of thenervous system as a nite interconnection of logical de-vices. For the rst time, neural networks were consid-ered as discrete abstract machines, and the issue of theircomputational capabilities investigated from the automata-theoretic perspective. Further developments of this per-spective opened up the way to the theoretical approach toneural computation [3, 4, 5].A Turingmachine (TM)consistsofainnitetape,aheadthat can read and write on this tape, and a nite programwhich, according to the current computational state of themachine and the current symbol read by the head, deter-mines the next symbol to be written by the head on thetape, the next move of the head (left or right), and the nextcomputational state of the machine. The classical Turingparadigmofcomputationcorrespondstothecomputationalscenario where a system receives a nite input, processesthis input, and either provides a corresponding output ornever halts. According to the Church-Turing Thesis, theTuring machine model is capable of capturing all possibleaspects of algorithmic computation [6].The concept of a Turing machine with advise (TM /A)provides a model of computation beyond the Turing lim-its. It consists of a classical Turing machine provided withan additional advise function : N ! f 0;1g

We introduce new algebro-topological invariants of directed networks, based on the topological construction of the directed clique complex. The shape of the underlying directed graph is encoded in a way that can be studied mathematically to obtain network invariants such as the Euler characteristic and the Betti numbers. Two different cases illustrate the application of the Euler characteristic. We investigate how the evolution of a Boolean recurrent artificial neural network is influenced by its topology in a dynamics involving pruning and strengthening of the connections, and to show that the topological features of the directed clique complex influence the dynamical evolution of the network. The second application considers the directed clique complex in a broader framework, to define an invariant of directed networks, the network degree invariant, which is constructed by computing the topological invariant on a sequence of sub-networks filtered by the minimum in- or out-degree of the nodes. The application of the Euler characteristic presented here can be extended to any directed network and provides a new method for the assessment of specific functional features associated with the network topology., Comment: 13 pages, 3 figures. Figures and captions improved, typographical changes

We present a deep convolutional neural network which is capable to distinguish between different contact states in robotic manipulation tasks. By integrating spatial and temporal tactile sensor data from a piezo-resistive sensor array through deep learning techniques, the network is not only able to classify the contact state into stable versus slipping, but also to distinguish between rotational and translation slippage. We evaluated different network layouts and reached a final classification rate of more than 97 %. Using consumer class GPUs, slippage and rotation events can be detected within 10 ms, which is still feasible for adaptive grasp control.

We study the attractor dynamics of a Boolean model of the basal ganglia-thalamocortical network as a function of its interactive synaptic connections and global threshold. We show that the regulation of the interactive feedback and global threshold are significantly involved in the maintenance and robustness of the attractor basin. These results support the hypothesis that, beyond mere structural architecture, global plasticity and interactivity play a crucial role in the computational and dynamical capabilities of biological neural networks.

The evolution of a simulated feed-forward neural network with recurrent excitatory connections and inhibitory forward connections is studied within the framework of algebraic topology. The dynamics includes pruning and strengthening of the excitatory connections. The invariants that we define are based on the connectivity structure of the underlying graph and its directed clique complex. The computation of this complex and of its Euler characteristic are related with the dynamical evolution of the network. As the network evolves dynamically, its network topology changes because of the pruning and strengthening of the onnections and algebraic topological invariants can be computed at different time steps providing a description of the process. We observe that the initial values of the topological invariant computed on the network before it evolves can predict the intensity of the activity.

We provide a decidable hierarchical classification of first-order recurrent neural networks made up of McCulloch and Pitts cells. This classification is achieved by proving an equivalence result between such neural networks and deterministic Büuchi automata, and then translating the Wadge classification theory from the abstract machine to the neural network context. The obtained hierarchy of neural networks is proved to have width 2 and height omega + 1, and a decidability procedure of this hierarchy is provided. Notably, this classification is shown to be intimately related to the attractive properties of the considered networks.

Precise spatiotemporal sequences of neuronal discharges (i.e., intervals between epochs repeating more often than expected by chance), have been observed in a large set of experimental electrophysiological recordings. Sensitivity to temporal information, by itself, does not demonstrate that dynamics embedded in spike trains can be transmitted through a neural network. This study analyzes how synaptic transmission through three archetypical types of neurons (regular-spiking, thalamo-cortical and resonator), simulated by a simple spiking model, can affect the transmission of precise timings generated by a nonlinear deterministic system (i.e., the Zaslavskii mapping in the present study). The results show that cells with subthreshold oscillations (resonators) are very sensitive to stochastic inputs, and are not a good candidate for transmitting temporally coded information. Thalamo-cortical neurons may transmit very well temporal patterns in the absence of background activity, but jitter accumulates along the synaptic chain. Conversely, we observed that cortical regular-spiking neurons can propagate filtered temporal information in a reliable way through the network, and with high temporal accuracy. We discuss the results in the general framework of neural dynamics and brain theories.

We consider a model of so-called hybrid recurrent neural networks composed with Boolean input and output cells as well as sigmoid internal cells. When subjected to some infinite binary input stream, the Boolean output cells necessarily exhibit some attractor dynamics, which is assumed to be of two possible kinds, namely either meaningful or spurious, and which underlies the arising of spatiotemporal patterns of output discharges. In this context, we show that rational-weighted neural networks are computationally equivalent to deterministic Muller Turing machines, whereas all other models of real-weighted or evolving neural networks are equivalent to each other, and strictly more powerful than deterministic Muller Turing machines. In this precise sense, the analog and evolving neural networks are super-Turing. We further provide some precise mathematical characterization of the expressive powers of all these neural models. These results constitute a generalization to the current computational context of those obtained in the cases of classical as well as interactive computations. They support the idea that recurrent neural networks represent a natural model of computation beyond the Turing limits.

The mesoscopic activity of the brain is strongly dynamical, while at the same time exhibits remarkable computational capabilities. In order to examine how these two features coexist, here we show that the patterns of synchronized oscillations displayed by networks of neural mass models, representing cortical columns, can be used as substrates for Boolean-like computations. Our results reveal that the same neural mass network may process different combinations of dynamical inputs as different logical operations or combinations of them. This dynamical feature of the network allows it to process complex inputs in a very sophisticated manner. The results are reproduced experimentally with electronic circuits of coupled Chua oscillators, showing the robustness of this kind of computation to the intrinsic noise and parameter mismatch of the coupled oscillators. We also show that the information-processing capabilities of coupled oscillations go beyond the simple juxtaposition of logic gates. Work supported by the Ministerio de Economia y Competividad (Spain) and FEDER through projects FIS2012-37655-C02-01 and FIS2013-41057-P. JG acknowledges support from the ICREA Academia programme and from the Generalitat de Catalunya (project 2014SGR0974)

Neurons in the brain are known to operate under a careful balance of excitation and inhibition, which maintains neural microcircuits within the proper operational range. How this balance is played out at the mesoscopic level of neuronal populations is, however, less clear. In order to address this issue, here we use a coupled neural mass model to study computationally the dynamics of a network of cortical macrocolumns operating in a partially synchronized, irregular regime. The topology of the network is heterogeneous, with a few of the nodes acting as connector hubs while the rest are relatively poorly connected. Our results show that in this type of mesoscopic network excitation and inhibition spontaneously segregate, with some columns acting mainly in an excitatory manner while some others have predominantly an inhibitory effect on their neighbors. We characterize the conditions under which this segregation arises, and relate the character of the different columns with their topological role within the network. In particular, we show that the connector hubs are preferentially inhibitory, the more so the larger the node's connectivity. These results suggest a potential mesoscale organization of the excitation-inhibition balance in brain networks., Author Summary One of the salient characteristics of neurons is their ability to either excite or inhibit other neurons, depending on the type of neurotransmitter they use to act upon them. In fact, a careful balance between excitation and inhibition is required for the brain to operate in a sustained state, away from both epileptic activity (which would arise if excitation dominated over inhibition) and complete quiescence (which would emerge if inhibition prevailed). While much work has been devoted to study how excitation and inhibition are organized at the level of neural microcircuits, little is known about how these two effects are structured at larger scales. Here we are interested in the scale of cortical macrocolumns, large collectives of neurons that can be described computationally by population models known as neural mass models. These models are routinely used to describe the brain rhythms observed with non-invasive techniques such as EEG. In this paper we show, using a neural mass model, that a collection of coupled cortical macrocolumns self-organizes spontaneously into a dynamical state in which excitation and inhibition are segregated at the mesoscopic level, with some cortical columns (the most connected ones) being inhibitory, and the rest being excitatory.

We present a complete overview of the computational power of recurrent neural networks involved in an interactive bio-inspired computational paradigm. More precisely, we recall the results stating that interactive rational- and realweighted neural networks are Turing-equivalent and super-Turing, respectively.We further prove that interactive evolving neural networks are super-Turing, irrespective of whether their synaptic weights are modeled by rational or real numbers. These results show that the computational powers of neural nets involved in a classical or in an interactive computational framework follow similar patterns of characterization. They suggest that some intrinsic computational capabilities of the brain might lie beyond the scope of Turing-equivalentmodels of computation, hence surpass the potentialities every current standard artificial models of computation.

One of the major obstacles found when trying to construct artefacts derived from principles observed in living beings is the lack of actual dynamic hardware with autonomous capabilities. Even if programmable devices offer the possibility of modifying the functionality implemented in the device, they rely on external hardware and software elements to provide its physical configuration. In this paper we present a new family of electronic devices, called POEtic, whose architecture has been derived from the basic properties that can be extracted from the three major organization principles present in living beings: phylogenesis, ontogenesis and epigenesis. We will demonstrate that the capabilities present in these new programmable devices make them an ideal candidate for the real-time emulation of large-scale biologically inspired spiking neural network models.

In this paper, we provide a historical survey of the most significant results concerning the computational power of neural models. We distinguish three important periods: first, the early works from McCulloch and Pitts, Kleene, and Minky, where the computational equivalence between Boolean recurrent neural networks and finite state automata is established. Secondly, the two breakthroughs by Siegelmann and Sontag showing the Turing universality of rational-weighted neural networks, and the super-Turing capabilities of analog recurrent neural networks. Thirdly, the recent results by Cabessa, Siegelmann and Villa revealing the super-Turing computational potentialities of interactive and evolving recurrent neural networks.

International audience; The present study demonstrates the application of the Unsupervised Spike Sorting algorithm (USS) to separation of multi-unit recordings and investigation of neuronal activity patterns in the subthalamic nucleus (STN). This nucleus is the main target for deep brain stimulation (DBS) in Parkinsonian patients. The USS comprises a fast unsupervised learning procedure and allows sorting of multiple single units, if any, out of a bioelectric signal. The algorithm was tested on a simulated signal with different levels of noise and with application of Time and Spatial Adaptation (TSA) algorithm for denoising. The results of the test showed a good quality of spike separation and allow its application to investigation of neuronal activity patterns in a medical application. One hundred twenty-four single channel multi-unit records from STN of 6 Parkinsonian patients were separated with USS into 492 single unit trains. Auto- and crosscorrellograms for each unit were analyzed in order to reveal oscillatory, bursting and synchronized activity patterns. We analyzed separately two brain hemispheres. For each hemisphere the percentage of units of each activity pattern were calculated. The results were compared for the first and the second operated hemispheres of each patient and in total.

We show that subsets of interacting oscillators may synchronize in different ways within a single network. This diversity of synchronization patterns is promoted by increasing the heterogeneous distribution of coupling weights and/or asymmetries in small networks. We also analyze consistency, defined as the persistence of coexistent synchronization patterns regardless of the initial conditions. Our results show that complex weighted networks display richer consistency than regular networks, suggesting why certain functional network topologies are often constructed when experimental data are analyzed., 9 pages, 6 figures

Classification with rejection is well understood for classifiers which provide explicit class probabilities. The situation is more complicated for popular deterministic classifiers such as learning vector quantisation schemes: albeit reject options using simple distance-based geometric measures were proposed [4], their local scaling behaviour is unclear for complex problems. Here, we propose a local threshold selection strategy which automatically adjusts suitable threshold values for reject options in prototype-based classifiers from given data. We compare this local threshold strategy to a global choice on artificial and benchmark data sets; we show that local thresholds enhance the classification results in comparison to global ones, and they better approximate optimal Bayesian rejection in cases where the latter is available.

The distribution of time intervals between successive spikes generated by a neuronal cell --the interspike intervals (ISI)-- may reveal interesting features of the underlying dynamics. In this study we analyze the ISI sequence --the spike train-- generated by a simple network of neurons whose output activity is modeled by a jump-diffusion process. We prove that, when specific ranges of the involved parameters are chosen, it is possible to observe multimodal ISI distributions which reveal that the modeled network fires with more than one single preferred time interval. Furthermore, the system exhibits resonance behavior, with modulation of the spike timings by the noise intensity. We also show that inhibition helps the signal transmission between the units of the simple network.

Understanding the dynamical and computational capabilities of neural models represents an issue of central importance. In this context, recent results show that interactive evolving recurrent neural networks are super-Turing, irrespective of whether their synaptic weights are rational or real. We extend these results by showing that interactive evolving recurrent neural networks are not only super-Turing, but also capable of simulating any other possible interactive deterministic system. In this sense, interactive evolving recurrent neural networks represents a super-Turing universal model of computation, irrespective of whether their synaptic weights are rational or real.

A critical feature of brain theories is whether neurons convey a noisy rate code or a precise temporal code. One of most valuable ways to test these theories consists in collecting the electrophysiological activity of cell assembles under several experimental conditions. The sequences of cell discharges—the spike trains—form time series whose dynamics is strongly related to the information processing carried out in the brain areas under study. Our purpose is to provide a user-friendly framework of a ‘Virtual Laboratory’ where computational neuroscience analyses and display of results can be distributed over a computer network, like Internet.

The existence of precise temporal relations in sequences of spike intervals, referred to as ‘spatiotemporal patterns’, is suggested by brain theories that emphasize the role of temporal coding. Specific analytical methods able to assess the significance of such patterned activity are extremely important to establish its function for information processing in the brain. This study proposes a new method called ‘pattern grouping algorithm’ (PGA), designed to identify and evaluate the statistical significance of patterns which differ from each other by a defined and small jitter in spike timing of the order of few ms. The algorithm performs a pre-selection of template patterns with a fast computational approach, optimizes the jitter for each spike in the template and evaluates the statistical significance of the pattern group using three complementary statistical approaches. Simulated data sets characterized by various types of known non stationarities are used for validation of PGA and for comparison of its performance to other methods. Applications of PGA to experimental data sets of simultaneously recorded spike trains are described in a companion paper (Tetko IV, Villa AEP. A pattern grouping algorithm for analysis of spatiotemporal patterns in neuronal spike trains. 2. Application to simultaneous single unit recordings. J Neurosci Methods 2000; accompanying article).

This study demonstrates the practical application of the pattern grouping algorithm (PGA), presented in the companion paper (Tetko IV, Villa AEP. A pattern grouping algorithm for analysis of spatiotemporal patterns in neuronal spike trains. 1. Detection of repeated patterns. J. Neurosci. Methods 2000; accompanying article), to data sets including up to 30 simultaneously recorded spike trains. The analysis of a large network of simulated neurons shows that the incidence of patterns cannot be simply related to an increase in firing rates obtained after Hebbian learning. Patterns that disappeared and reappeared in the thalamus of anesthetized rats when the cerebral cortex was reversibly inactivated suggest that widespread cell assemblies contribute to the generation and propagation of precisely timed activity. In an another experiment multiple spike trains were recorded from the temporal cortex of freely moving rats performing a complex two-choice discrimination task. The presence or absence of particular patterns in the period preceding the cue was associated with changes in reaction time. In conclusion, neuronal network interactions may generate spatiotemporal firing patterns detectable by PGA. We provide evidence of such patterned activity associated with specific animal's behavior, thus suggesting the existence of complex temporal coding schemes in the higher nervous centers of the brain.

This article presents a self-organising multilayered iterative algorithm that provides linear and non-linear polynomial regression models thus allowing the user to control the number and the power of the terms in the models. The accuracy of the algorithm is compared to the partial least squares (PLS) algorithm using fourteen data sets in quantitative-structure activity relationship studies. The calculated data show that the proposed method is able to select simple models characterized by a high prediction ability and thus provides a considerable interest in quantitative-structure activity relationship studies. The software is developed using client-server protocol (Java and C++ languages) and is available for world-wide users on the Web site of the authors.

Pruning methods for feed-forward artificial neural networks trained by the cascade-correlation learning algorithm are proposed. The cascade-correlation algorithm starts with a small network and dyn...

The present study investigates an application of artificial neural networks (ANNs) for use in pharmaceutical fingerprinting. Several pruning algorithms were applied to decrease the dimension of the input parameter data set. A localized fingerprint region was identified within the original input parameter space from which a subset of input parameters was extracted leading to enhanced ANN performance. The present results confirm that ANNs can provide a fast, accurate, and consistent methodology applicable to pharmaceutical fingerprinting.

This study investigates the emerging possibilities of combining unsupervised and supervised learning in neural network ensembles. Such strategy is used to get an efficient partition of a noisy input data set in order to focus the training of neural networks on the most complex and informative domains of the data set and accelerate the learning phase. The proposed algorithm provides a good prediction accuracy using fewer cases from non-informative domains according to a correlative measure of dependency between cases of the training set. This measure takes into account internal relationships amid analyzed data and can be used to cluster neighbor cases in a multidimensional space and to filter out the outliers. The possible relation of the proposed algorithm to brain processing occurring in the thalamo-cortical pathway is discussed.

This study presents two efficient algorithms – combinatorial and probabilistic combinatorial methods (CM and PCM) – for estimation of a number of precise patterns of discharges that occur by chance in records of multiple single-unit spike trains. The confidence limits estimated by these methods are in good agreement with different sets of simulated test data as well as with the ad-hoc method. Both combinatorial methods provided a better accuracy than the bootstrap algorithm and in most cases of nonstationary data PCM provided better estimations than the ad-hoc method. Introduction of a jitter for searching patterns with a precision of a few milliseconds and burst filtering may introduce biases in the estimations. Comparison of a new filtering procedure based upon a filtering frequency with previously described schemes of filtering indicates the possibility of using a simple setting which remains accurate over a wide range of parameters. We aim to implement a combination of PCM for estimations of the number of patterns formed by three to seven spikes and CM for higher-order complexities for estimations during experiments in progress.

We provide a novel refined attractor-based complexity measurement for Boolean recurrent neural networks that represents an assessment of their computational power in terms of the significance of their attractor dynamics. This complexity measurement is achieved by first proving a computational equivalence between Boolean recurrent neural networks and some specific class of ω-automata, and then translating the most refined classification of ω-automata to the Boolean neural network context. As a result, a hierarchical classification of Boolean neural networks based on their attractive dynamics is obtained, thus providing a novel refined attractor-based complexity measurement for Boolean recurrent neural networks. These results provide new theoretical insights to the computational and dynamical capabilities of neural networks according to their attractive potentialities. An application of our findings is illustrated by the analysis of the dynamics of a simplified model of the basal ganglia-thalamocortical network simulated by a Boolean recurrent neural network. This example shows the significance of measuring network complexity, and how our results bear new founding elements for the understanding of the complexity of real brain circuits.

Understanding the dynamical and computational capabilities of neural models represents an issue of central importance. Here, we consider a model of first-order recurrent neural networks provided with the possibility to evolve over time and involved in a basic interactive and memory active computational paradigm. In this context, we prove that the so-called interactive evolving recurrent neural networks are computationally equivalent to interactive Turing machines with advice, hence capable of super-Turing potentialities. We further provide a precise characterisation of the ω-translations realised by these networks. Therefore, the consideration of evolving capabilities in a first-order neural model provides the potentiality to break the Turing barrier.

A multiple layers neural network characterized by diverging/converging projections between successive layers activated by an external spatio-temporal input pattern in presence of stochastic background activities was considered. In the previous studies we reported the properties and performance of spike information transmission in the network depending on neuron model type, inputed information type and background activity level. The models were rather simple and can be more detailed and bigger in size for further investigation. Based on a technology developed in the integrated physiology, we have implemented the network model on PhysioDesigner, a platform for multilevel mathematical modeling of physiological systems. This article instructs a use case of PhysioDesigner and the assistive function of PhysioDesigner especially for large size neuronal network modeling is demonstrated.

We investigated the effect of reciprocal connections in a network of modules of simulated spiking neurons. The neural activity is recorded by means of virtual electrodes and EEG-like signals, called electrochipograms (EChG), are analyzed by time- and frequency-domain methods. Bio-inspired processes in the circuits drive the build-up of auto-associative links within each module, which generate an areal activity, recorded by EChG, that reflect the changes in the corresponding functional connectivity within and between neuronal modules. We found that circuits with short inter-layer reciprocal projections exhibited enhanced response as to the stimulus, as to the inner-activity and long inter-layer projections make circuit exhibit non-coherent behavior. We show evidence that all networks of modules are able to process and maintain patterns of activity associated with the stimulus after its offset.

International audience; Modular architecture is a hallmark of many brain circuits. In the cerebral cortex, in particular, it has been observed that reciprocal connections are often present between functionally interconnected areas that are hierarchically organized. We investigate the effect of reciprocal connections in a network of modules of simulated spiking neurons. The neural activity is recorded by means of virtual electrodes and EEG-like signals, called electrochipograms (EChG), analyzed by time- and frequency-domain methods. A major feature of our approach is the implementation of important bio-inspired processes that affect the connectivity within a neural module: synaptogenesis, cell death, spike-timing-dependent plasticity and synaptic pruning. These bio-inspired processes drive the build-up of auto-associative links within each module, which generate an areal activity, recorded by EChG, that reflect the changes in the corresponding functional connectivity within and between neuronal modules. We found that circuits with intra-layer reciprocal projections exhibited enhanced stimulus-locked response. We show evidence that all networks of modules are able to process and maintain patterns of activity associated with the stimulus after its offset. The presence of feedback and horizontal projections was necessary to evoke cross-layer coherence in bursts of -frequency at regular intervals. These findings bring new insights to the understanding of the relation between the functional organization of neural circuits and the electrophysiological signals generated by large cell assemblies. This article is part of a Special Issue entitled "Neural Coding".

A ten layers feedforward network characterized by diverging/ converging patterns of projection between successive layers is activated by an external spatio-temporal input pattern fed to layer 1 in presence of stochastic background activities fed to all layers. We used three dynamical systems to derive the external input spike trains including the temporal information, and two types of neuron models for the network, i.e. either a simple spiking neuron (SSN) or a multiple-timescale adaptive threshold neuron (MAT). We observed an unimodal integration effect as a function of the order of the layers and confirmed that the MAT model is likely to be more efficient in integrating and transmitting the temporal structure embedded in the external input.

We simulated the activity of hierarchically organized spiking neural networks characterized by an initial developmental phase featuring cell death followed by spike timing dependent synaptic plasticity in presence of background noise. Upstream networks receiving spatiotemporally organized external inputs projected to downstream networks disconnected from external inputs. The observation of precise firing sequences, formed by recurrent patterns of spikes intervals above chance levels, suggested the build-up of an unsupervised connectivity able to sustain and preserve temporal information processing.

OpenAdap.net (OAN) is an information system aimed at knowledge dissemination and processing for sustaining communities of users who share the same knowledge representation within their field of interest. Such information system is not only the technology a community uses, but also the way in which the members of the community interact with the technology and the way in which the technology processes knowledge within the community. OpenAdap.net provides multiple business opportunities through the introduction of an innovative distributed working environment and creative user collaborations. In the framework, OAN-aware applications are established software pieces, open or closed sourced, developed to their own independent goals, which are extended to access the shared resources through the use of OAN-specific code.

International audience; Complex neural modules with embedded neural development and synaptic plasticity features have been connected to form a hierarchical recurrent circuit. Virtual electrodes have been used to record a "neural" generated signal, called electrochipogram EChG, from each module. The EChG are processed by frequency domain methods to determine the modifications in functional connectivity by assessing quadratic phase coupling. The experimental paradigm is aimed to describe what happened prior to, at the beginning, towards the end, and after repeating an external input at fixed frequency. The results are discussed by comparing with the same signal processing methods applied to a human study.

This study investigates the ability of a diverging/converging neural network to transmit and integrate a complex temporally organized activity embedded in afferent spike trains. The temporal information is originally generated by a deterministic nonlinear dynamical system whose parameters determine a chaotic attractor. We present the simulations obtained with a network formed by simple spiking neurons (SSN) and a network formed by a multiple-timescale adaptive threshold neurons (MAT). The assessment of the temporal structure embedded in the spike trains is carried out by sorting the preferred firing sequences detected by the pattern grouping algorithm (PGA). The results suggest that adaptive threshold neurons are much more efficient in maintaining a specific temporal structure distributed across multiple spike trains throughout the layers of a feed-forward network.