| Presentation | 1997/5/23 Unsupervised Learning of Data Correction : What Do Sandglass Networks Learn? Kazuyuki HIRAOKA, Shuji Yoshizawa, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | We treat a problem to estimate a relation among true values from noisy data, and correct the data using the estimated relation. It is also regarded as a learning problem for the case that both inputs and outputs of examples are noisy (errors-in-variables model). The equivalence between the maximum likelihood estimate for this problem and the autoassociative learning of the sandglass network is pointed out. The sandglass network can treat multi-valued functions. However, it is able to express only surfaces (manifolds) which are homeomorphic to the plane (hyperplane). Thus the learning of closed surfaces like the sphere is hard for it. We multiply the sandglass network and construct the mixture-of-experts type network, which is able to express a broad class of multi-valued functions. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | sandglass network / data correction / autoassociative learning / errors-in-variables model / mixture of experts / nonlinear principal components analysis |
| Paper # | NC97-4 |
| Date of Issue |
| Conference Information | |
| Committee | NC |
|---|---|
| Conference Date | 1997/5/23(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) | Unsupervised Learning of Data Correction : What Do Sandglass Networks Learn? |
| Sub Title (in English) | |
| Keyword(1) | sandglass network |
| Keyword(2) | data correction |
| Keyword(3) | autoassociative learning |
| Keyword(4) | errors-in-variables model |
| Keyword(5) | mixture of experts |
| Keyword(6) | nonlinear principal components analysis |
| 1st Author's Name | Kazuyuki HIRAOKA |
| 1st Author's Affiliation | Department of Information Engineering, University of Tokyo() |
| 2nd Author's Name | Shuji Yoshizawa |
| 2nd Author's Affiliation | Department of Information Engineering, University of Tokyo |
| Date | 1997/5/23 |
| Paper # | NC97-4 |
| Volume (vol) | vol.97 |
| Number (no) | 69 |
| Page | pp.pp.- |
| #Pages | 8 |
| Date of Issue | |