Presentation | 2005/7/20 A Method of Active Learning with Model Selection Sami HANHIJARVI, Masashi SUGIYAMA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | In active learning, optimal input points are selected for a single fixed model. Conversely, in model selection, several models are trained with a given fixed set of examples and the best model is chosen. Simultaneously solving these problems, a problem called active learning with model selection, cannot be accomplished by simply combining conventional active learning and model selection methods in a batch manner. In this paper, we propose two sequential methods for iteratively choosing a model and an input point. In the first method, the most promising model is chosen at each step and the next input point is optimized for that model. This method is expected to work well if a reasonably good model is chosen in an early stage of the sequential learning process. On the other hand, good input points for one model could be poor for others. This implies that the chosen input points could be overfitted to the chosen model. Therefore, if the model chosen in an early stage is found to be not promising and not selected in the end, the method may not yield desired results. In order to alleviate this overfitting problem caused by the 'hard' selection of the model, we propose a 'soft' selection strategy, i.e., the next input point is chosen not for the single best model but for several good models. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Active learning / model selection / active learning with model selection / generalization error / covariate shift |
Paper # | NC2005-36 |
Date of Issue |
Conference Information | |
Committee | NC |
---|---|
Conference Date | 2005/7/20(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 | ENG |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A Method of Active Learning with Model Selection |
Sub Title (in English) | |
Keyword(1) | Active learning |
Keyword(2) | model selection |
Keyword(3) | active learning with model selection |
Keyword(4) | generalization error |
Keyword(5) | covariate shift |
1st Author's Name | Sami HANHIJARVI |
1st Author's Affiliation | Laboratory of Computer and Information Science, Helsinki University of Technology:Department of Computer Science, Tokyo Institute of Technology() |
2nd Author's Name | Masashi SUGIYAMA |
2nd Author's Affiliation | Department of Computer Science, Tokyo Institute of Technology |
Date | 2005/7/20 |
Paper # | NC2005-36 |
Volume (vol) | vol.105 |
Number (no) | 211 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |