| Presentation | 1999/6/18 Adaptive algorithm for active learning using an information criterion for the maximum weighted log-likelihood estimator Takafumi Kanamori, Hidetoshi Shimodaira, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | We suppose that outputs of a system is expressed as probability conditioned by inputs. In this paper we study the estimation of the system when the observer can select appropriate inputs to the system. We call such method of estimation Active Learning. We suppose that the statistical model does not include the system. When the statistical model is not correct, the maximum likelihood estimator does not have the property of consistency. Here consistency means the convergence to the optimal parameter measured by the Kullback-Leibler divergence. Hence we suggest an active learning algorithm using maximum weighted log-likelihood estimator (mwle). The algorithm has the property of consistency. Moreover we point out that there are estimators which are better than the consistent estimators when the number of the data is finite. Considering such result we construct another active learning algorithm which select appropriate mwle using information criterion. We do computer simulation and consider the effct of the suggested algorithms. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Active learning / Maximum weighted log-likelihood estimator / Statistical risk / Information criterion / Statistical asymptotic theory |
| Paper # | NC99-12 |
| Date of Issue |
| Conference Information | |
| Committee | NC |
|---|---|
| Conference Date | 1999/6/18(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) | Adaptive algorithm for active learning using an information criterion for the maximum weighted log-likelihood estimator |
| Sub Title (in English) | |
| Keyword(1) | Active learning |
| Keyword(2) | Maximum weighted log-likelihood estimator |
| Keyword(3) | Statistical risk |
| Keyword(4) | Information criterion |
| Keyword(5) | Statistical asymptotic theory |
| 1st Author's Name | Takafumi Kanamori |
| 1st Author's Affiliation | The Graduate University for Advanced Studies() |
| 2nd Author's Name | Hidetoshi Shimodaira |
| 2nd Author's Affiliation | The Institute of Statistical Mathematics |
| Date | 1999/6/18 |
| Paper # | NC99-12 |
| Volume (vol) | vol.99 |
| Number (no) | 131 |
| Page | pp.pp.- |
| #Pages | 8 |
| Date of Issue | |