Exploration of Mean-Field Approximation for Feed-Forward Networks - Neural Networks, 1999. IJCNN '99. International Joint Conference on
نویسنده
چکیده
We present a formulation of mean-field approximation for layered feed-forward stochastic networks. In this formulation, one can obtain not only estimates of averages for state variables of the networks but also those of intra-layer correlations, the latter of which cannot be obtained by the conventional mean-jield approximation. Moreovel; this formulation provides a pamework to treat “conditional ’’ expectations, expectations under the constraint that external information about statistics are fed to some layers of the network, which plays an important role in several applications such as the Helmholtz machine.
منابع مشابه
Exploration of mean-field approximation for feedforward networks
We present a formulation of mean-field approximation for layered feed-forward stochastic networks. In this formulation, one can obtain not only estimates of averages for state variables of the networks but also those of intra-layer correlations, the latter of which cannot be obtained by the conventional mean-field approximation. Moreover, this formulation provides a framework to treat “conditio...
متن کاملSynchronized chaos in coupled neuromodules of different types
We discuss the time-discrete parametrized dynamics of two coupled recurrent neural networks. General conditions for the existence of synchronized dynamics are derived for these systems, and it is demonstrated that also the coupling of totally different network structures can result in periodic, quasiperiodic as well as chaotic dynamics constrained to a synchronization manifold M . Stability of ...
متن کاملSTRUCTURAL DAMAGE DETECTION BY MODEL UPDATING METHOD BASED ON CASCADE FEED-FORWARD NEURAL NETWORK AS AN EFFICIENT APPROXIMATION MECHANISM
Vibration based techniques of structural damage detection using model updating method, are computationally expensive for large-scale structures. In this study, after locating precisely the eventual damage of a structure using modal strain energy based index (MSEBI), To efficiently reduce the computational cost of model updating during the optimization process of damage severity detection, the M...
متن کاملA Boundary-pair Representation for Perception Modeling - Neural Networks, 1999. IJCNN '99. International Joint Conference on
It is widely accepted that responses from onand off-center cells give rise to edges and are equivalent to edge detectors. In this paper, we point out that onand off-center cell responses provide more information than edges. W e show that an edge-based representation makes the ownership of boundaries ambiguous and requires a combinatorial search to model perceptual grouping. By analyzing the dif...
متن کاملPREDICTION OF COMPRESSIVE STRENGTH AND DURABILITY OF HIGH PERFORMANCE CONCRETE BY ARTIFICIAL NEURAL NETWORKS
Neural networks have recently been widely used to model some of the human activities in many areas of civil engineering applications. In the present paper, artificial neural networks (ANN) for predicting compressive strength of cubes and durability of concrete containing metakaolin with fly ash and silica fume with fly ash are developed at the age of 3, 7, 28, 56 and 90 days. For building these...
متن کامل