The authors propose a new algorithm which builds feedforward layered network in order to learn any Boolean function of N units. number layers and the hidden units each layer are not prescribed advance: they outputs algorithm. It is an for growth network, adds layers, inside layer, at will until convergence. convergence guaranteed numerical tests this strategy look promising.

Abstract We present a model of opinion dynamics in which agents adjust continuous opinions as result random binary encounters whenever their difference is below given threshold. High thresholds yield convergence toward an average opinion, whereas low several clusters. The further generalized to network interactions, threshold heterogeneity, adaptive thresholds, and strings opinions. © 2002 Wiley Periodicals, Inc.

The increasing integration of technology into our lives has created unprecedented volumes data on society's everyday behaviour. Such opens up exciting new opportunities to work towards a quantitative understanding complex social systems, within the realms discipline known as Computational Social Science. Against background financial crises, riots and international epidemics, urgent need for greater comprehension complexity interconnected global society an ability apply such insights in...

In the context of parameter estimation and model selection, it is only quite recently that a direct link between Fisher information information-theoretic quantities has been exhibited. We give an interpretation this within standard framework theory. show in population coding, mutual activity large array neurons stimulus to which are tuned naturally related information. light result, we consider optimization tuning curves parameters case responding represented by angular variable.

A model for formal neural networks that learn temporal sequences by selection is proposed on the basis of observations acquisition song birds, sequence-detecting neurons, and allosteric receptors. The relies hypothetical elementary devices made up three synaptic triads, which yield short-term modification efficacy through heterosynaptic interactions, a local Hebbian learning rule. functional units postulated are mutually inhibiting clusters synergic neurons bundles synapses. Networks...

We investigate the consequences of maximizing information transfer in a simple neural network (one input layer, one output layer), focusing on case nonlinear functions. assume that both receptive fields (synaptic efficacies) and functions can be adapted to environment. The main result is that, for bounded invertible functions, vanishing additive noise, no maximization (Linsker's infomax principle) leads factorial code-hence same solution as required by redundancy-reduction principle Barlow....

One characteristic behaviour of the Hopfield model neural networks, namely catastrophic deterioration memory due to overloading, is interpreted in simple physical terms. A general formulation allows for an exploration some basic issues learning theory. Two schemes are constructed, which avoid overloading and keep forgetting, with a stationary capacity.

This letter suggests that in biological organisms, the perceived structure of reality, particular notions body, environment, space, object, and attribute, could be a consequence an effort on part brains to account for dependency between their inputs outputs terms small number parameters. To validate this idea, procedure is demonstrated whereby brain (simulated) organism with arbitrary input output connectivity can deduce dimensionality rigid group space underlying its input-output...

We investigate the consequences of maximizing information transfer in a simple neural network (one input layer, one output layer), focusing on case nonlinear functions. assume that both receptive fields (synaptic efficacies) and functions can be adapted to environment. The main result is that, for bounded invertible functions, vanishing additive noise, no maximization (Linsker's infomax principle) leads factorial code-hence same solution as required by redundancy-reduction principle Barlow....

2014 Nous considérons une famille de modèles qui généralise le modèle Hopfield, et peut s'étudier façon analogue.Cette englobe des schémas type palimpseste, dont les propriétés s'apparentent à celles d'une mémoire travail (mémoire court terme).En utilisant la méthode répliques, nous obtenons un formalisme simple permet comparaison détaillée divers d'apprentissage, l'étude d'effets variés, tel l'apprentissage par répétition.Abstract.2014 We consider a family of models, which generalizes the...

Proves a conjecture giving the exact number of directed animals s sites with any root, on strip finite width square lattice. The authors also rederive more simply some previous results concerning connective constant and particular eigenvectors transfer matrix.

The authors propose a new classifier based on neural network techniques. 'network' consists of set perceptrons functionally organized in binary tree ('neural tree'). learning algorithm is inspired from growth algorithm, the tiling recently introduced for feedforward networks. As former case, this constructive which convergence guaranteed. In one distinguishes structural organization functional organization: each neuron receives inputs from, and only input layer; its output does not feed into...

Abstract As a large-scale instance of dramatic collective behaviour, the 2005 French riots started in poor suburb Paris, then spread all France, lasting about three weeks. Remarkably, although there were no displacements rioters, riot activity did travel. Access to daily national police data has allowed us explore dynamics propagation. Here we show that an epidemic-like model, with just few parameters and single sociological variable characterizing neighbourhood deprivation, accounts...

We study an algorithm for a feedforward network which is similar in spirit to the Tiling recently introduced: hidden units are added one by until performs desired task, and convergence guaranteed. The difference architecture of network, more constrained here. Numerical tests show performances that algorithm, although total number couplings general grows faster.

We study simple, feedforward, neural networks for pattern storage and retrieval, with information theory criteria. Two Hebbian learning rules are considered, emphasis on sparsely coded patterns. address the question: under which conditions is optimal reached in error-full regime?For model introduced some time ago by Willshaw, Buneman Longuet-Higgins, stored goes through a maximum, may be found within error-less or regimes according to value of coding rate. However, it eventually vanishes as...

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not.The documents may come from teaching institutions in France abroad, public private centers.L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de scientifiques niveau recherche, publiés ou non, émanant des établissements d'enseignement recherche français étrangers, laboratoires publics privés. Directed lattice...

The cerebellum aids the learning of fast, coordinated movements. According to current consensus, erroneously active parallel fibre synapses are depressed by complex spikes signalling movement errors. However, this theory cannot solve

We show that the statistics of an edge type variable in natural images exhibits self-similarity properties which resemble those local energy dissipation turbulent flows. Our results and extended hold remarkably for variance, very same models can be used to predict all associated exponents. These suggest using as a laboratory testing more elaborate scaling interest statistical description The we have exhibited are relevant modeling early visual system: They should included designed prediction...

The storage and retrieval of complex sequences, with bifurcation points, for instance, in fully connected networks formal neurons, is investigated. We present a model which involves the transmission informations undergoing various delays from all neurons to one neuron, through synaptic connections, possibly high order. Assuming parallel dynamics, an exact solution proposed; it allows store without errors number elementary transitions are order connections related neuron. A fast-learning...

We propose an agent-based model of a single-asset financial market, described in terms small number parameters, which generates price returns with statistical properties similar to the stylized facts observed time series. Our generically leads absence autocorrelation returns, self-sustaining excess volatility, mean-reverting volatility clustering and endogenous bursts market activity non-attributable external noise. The parsimonious structure allows identification feedback heterogeneity as...