Metamorphosis Networks: An Alternative to Constructive Models
نویسندگان
چکیده
Given a set oft raining examples, determining the appropriate number of free parameters is a challenging problem. Constructive learning algorithms attempt to solve this problem automatically by adding hidden units, and therefore free parameters, during learning. We explore an alternative class of algorithms-called metamorphosis algorithms-in which the number of units is fixed, but the number of free parameters gradually increases during learning. The architecture we investigate is composed of RBF units on a lattice, which imposes flexible constraints on the parameters of the network. Virtues of this approach include variable subset selection, robust parameter selection, multiresolution processing, and interpolation of sparse training data.
منابع مشابه
Metamorphosis Networks: An Alternative to Constructive Methods
Given a set of training examples, determining the appropriate number of free parameters is a challenging problem. Constructive learning algorithms attempt to solve this problem automatically by adding hidden units, and therefore free parameters, during learning. We explore an alternative class of algorithms|called metamorphosis algorithms|in which the number of units is xed, but the number of f...
متن کاملMonthly runoff forecasting by means of artificial neural networks (ANNs)
Over the last decade or so, artificial neural networks (ANNs) have become one of the most promising tools formodelling hydrological processes such as rainfall runoff processes. However, the employment of a single model doesnot seem to be an appropriate approach for modelling such a complex, nonlinear, and discontinuous process thatvaries in space and time. For this reason, this study aims at de...
متن کاملRainfall-runoff modelling using artificial neural networks (ANNs): modelling and understanding
In recent years, artificial neural networks (ANNs) have become one of the most promising tools in order to model complex hydrological processes such as the rainfall-runoff process. In many studies, ANNs have demonstrated superior results compared to alternative methods. ANNs are able to map underlying relationship between input and output data without prior understanding of the process under in...
متن کاملConstructive Hypervolume Modeling Using Extended Space Mappings
In this paper, we propose a framework for modeling and deforming heterogeneous volumetric objects defined as point sets with attributes. We propose to use constructive hypervolume objects, where the function representation (FRep) is used as the basic model for both object geometry and attributes represented independently using realvalued scalar functions of point coordinates. While FRep directl...
متن کاملConstrained consumable resource allocation in alternative stochastic networks via multi-objective decision making
Many real projects complete through the realization of one and only one path of various possible network paths. Here, these networks are called alternative stochastic networks (ASNs). It is supposed that the nodes of considered network are probabilistic with exclusive-or receiver and exclusive-or emitter. First, an analytical approach is proposed to simplify the structure of t...
متن کامل