| Presentation | 2012-12-12 A numerical derivation of learning coefficient in radial basis function network Satoru Tokuda, Kenji Nagata, Masato Okada, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Radial basis function (RBF) network is a regression model which regresses input-output data by radial basis functions such as Gaussian function. Recently, this model has been widely used to the spectral deconvolution such as X-ray Photoelectron Spectroscopy data analysis, and enables us to estimate electronic state of matter from the peak positions of the estimated peaks. For the model selection for the RBF network, the well-known information criterion such as AIC and BIC cannot apply because the asymptotic normality does not hold due to the hierarchy of this model. In a recent study, new information criterion called WBIC has been proposed for the model selection of the models with hierarchy. WBIC, however, requires the numerical value of learning coefficient, which is the coefficient of the leading term for the free energy. In this paper, we propose a new method for calculating the learning coefficient, and derivate the coefficient in RBF network. More concretly, we use the exchange Monte Carlo method, and calculate the learning coefficient from the exchange ratio given by the result of the simulation. In addition, we simulate the learning in the variance-fixed-type RBF network, whose learning coefficient is clarified, and show the effectiveness of the proposed method by comparing the theoretical values and the experimental ones. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Radial basis function network / Learning coefficient / Exchange Monte Carlo method / Exchange ratio |
| Paper # | NC2012-78 |
| Date of Issue |
| Conference Information | |
| Committee | NC |
|---|---|
| Conference Date | 2012/12/5(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Neurocomputing (NC) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A numerical derivation of learning coefficient in radial basis function network |
| Sub Title (in English) | |
| Keyword(1) | Radial basis function network |
| Keyword(2) | Learning coefficient |
| Keyword(3) | Exchange Monte Carlo method |
| Keyword(4) | Exchange ratio |
| 1st Author's Name | Satoru Tokuda |
| 1st Author's Affiliation | The University of Tokyo() |
| 2nd Author's Name | Kenji Nagata |
| 2nd Author's Affiliation | The University of Tokyo |
| 3rd Author's Name | Masato Okada |
| 3rd Author's Affiliation | The University of Tokyo |
| Date | 2012-12-12 |
| Paper # | NC2012-78 |
| Volume (vol) | vol.112 |
| Number (no) | 345 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |