Résultats de recherche pour « Neural Network Models »

Many underlying relationships among data in several areas of science and engineering, e.g., computer vision, molecular chemistry, molecular biology, pattern recognition, and data mining, can be represented in terms of graphs. In this paper, we propose a new neural network model, called graph neural network (GNN) model, that extends existing neural network methods for processing the data represented in graph domains. This GNN model, which can directly process most of the practically useful types of graphs, e.g., acyclic, cyclic, directed, and undirected, implements a function tau(G,n) is an element of IR(m) that maps a graph G and one of its nodes n into an m-dimensional Euclidean space. A supervised learning algorithm is derived to estimate the parameters of the proposed GNN model. The computational cost of the proposed algorithm is also considered. Some experimental results are shown to validate the proposed learning algorithm, and to demonstrate its generalization capabilities.

A neural network model was developed to analyze and forecast the behavior of the river Tagliamento, in Italy, during heavy rain periods. The model makes use of distributed rainfall information coming from several rain gauges in the mountain district and predicts the water level of the river at the section closing the mountain district. The water level at the closing section in the hours preceding the event was used to characterize the behavior of the river system subject to the rainfall perturbation. Model predictions are very accurate (i.e., mean square error is less than 4%) when the model is used with a 1‐hour time horizon. Increasing the time horizon, thus making the model suitable for flood forecasting, decreases the accuracy of the model. A limiting time horizon is found corresponding to the minimum time lag between the water level at the closing section and the rainfall, which is characteristic of each flooding event and depends on the rainfall and on the state of saturation of the basin. Performance of the model remains satisfactory up to 5 hours. A model of this type using just rainfall and water level information does not appear to be capable of predicting beyond this time limit.

The author reviews recent developments in neural networks that are of general interest to physicists engaged in biological and biomedical research. Examples are examined of models streamlined for collective computation (McCulloch-Pitts, Little, Hopfield and Cowan-Hopfield models), Neural networks in theoretical neurobiology and synthetic neural networks are also detailed.

Over the past few years, neural networks have re-emerged as powerful machine-learning models, yielding state-of-the-art results in fields such as image recognition and speech processing. More recently, neural network models started to be applied also to textual natural language signals, again with very promising results. This tutorial surveys neural network models from the perspective of natural language processing research, in an attempt to bring natural-language researchers up to speed with the neural techniques. The tutorial covers input encoding for natural language tasks, feed-forward networks, convolutional networks, recurrent networks and recursive networks, as well as the computation graph abstraction for automatic gradient computation.

We present a neural network model that aims to bridge the historical gap between dynamic and structural approaches to personality. The model integrates work on the structure of the trait lexicon, the neurobiology of personality, temperament, goal-based models of personality, and an evolutionary analysis of motives. It is organized in terms of two overarching motivational systems, an approach and an avoidance system, as well as a general disinhibition and constraint system. Each overarching motivational system influences more specific motives. Traits are modeled in terms of differences in the sensitivities of the motivational systems, the baseline activation of specific motives, and inhibitory strength. The result is a motive-based neural network model of personality based on research about the structure and neurobiology of human personality. The model provides an account of personality dynamics and person-situation interactions and suggests how dynamic processing approaches and dispositional, structural approaches can be integrated in a common framework.

The typical fraction of the space of interactions between each pair of N Ising spins which solve the problem of storing a given set of p random patterns as N-bit spin configurations is considered. The volume is calculated explicitly as a function of the storage ratio, alpha =p/N, of the value kappa (>0) of the product of the spin and the magnetic field at each site and of the magnetisation, m. Here m may vary between 0 (no correlation) and 1 (completely correlated). The capacity increases with the correlation between patterns from alpha =2 for correlated patterns with kappa =0 and tends to infinity as m tends to 1. The calculations use a saddle-point method and the order parameters at the saddle point are assumed to be replica symmetric. This solution is shown to be locally stable. A local iterative learning algorithm for updating the interactions is given which will converge to a solution of given kappa provided such solutions exist.

Several neural network models in continuous time are reconsidered in the framework of a general mean-field theory which is exact in the limit of a large and fully connected network. The theory assumes pointlike spikes which are generated by a renewal process. The effect of spikes on a receiving neuron is described by a linear response kernel which is the dominant term in a weak-coupling expansion. It is shown that the resulting ``spike response model'' is the most general renewal model with linear inputs. The standard integrate-and-fire model forms a special case. In a network structure with several pools of identical spiking neurons, the global states and the dynamic evolution are determined by a nonlinear integral equation which describes the effective interaction within and between different pools. We derive explicit stability criteria for stationary (incoherent) and oscillatory (coherent) solutions. It is shown that the stationary state of noiseless systems is ``almost always'' unstable. Noise suppresses fast oscillations and stabilizes the system. Furthermore, collective oscillations are stable only if the firing occurs while the synaptic potential is increasing. In particular, collective oscillations in a network with delayless excitatory interaction are at most semistable. Inhibitory interactions with short delays or excitatory interactions with long delays lead to stable oscillations. Our general results allow a straightforward application to different network models with spiking neurons. Furthermore, the theory allows an estimation of the errors introduced in firing rate or ``graded-response'' models.