| Presentation | 2011-07-25 Importance weighted incremental learning of a limited kernel machine Koichiro YAMAUCHI, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | The author had already presented a kernel machine for embedded systems, namely, Limited General Regression Neural Networks, whose number of kernels is limited to a certain number. The fundamental architecture of the learning machine is the same as the General Regression Neural Networks. Initially, the LGRNN learns new samples incrementally by allocating new hidden units. If the number of hidden units reaches the upper bound, the LGRNN has to remove one useless hidden unit to learn a new sample. However, there are cases in which the adverse effects of removing a useless unit are greater than the positive effects of learning the new sample. In this case, the LGRNN should refrain from learning the new sample. To achieve this, the LGRNN predicts the effects of several learning options (e.g., ignore or learning) before the learning process begins, and chooses the best learning option to be executed. On the other hand, there are cases where the LGRNN fails to get precise selection of the learning option to be executed when the prior distribution of input is changing drastically. To overcome this problem, a new LGRNN proposed here introduces an importance weight for each learning sample. The importance weight is calculated based on the novelty of the sample, a Experimental results show that the method successfully reduces errors even when the prior distribution is changed drastically. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Kernel Method / Limited General Regression Neural Networks (LGRNN) / Incremental Learning / Approximated Linear Dependency / Pruning with replacement / importance weight |
| Paper # | NC2011-28 |
| Date of Issue |
| Conference Information | |
| Committee | NC |
|---|---|
| Conference Date | 2011/7/18(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) | Importance weighted incremental learning of a limited kernel machine |
| Sub Title (in English) | |
| Keyword(1) | Kernel Method |
| Keyword(2) | Limited General Regression Neural Networks (LGRNN) |
| Keyword(3) | Incremental Learning |
| Keyword(4) | Approximated Linear Dependency |
| Keyword(5) | Pruning with replacement |
| Keyword(6) | importance weight |
| 1st Author's Name | Koichiro YAMAUCHI |
| 1st Author's Affiliation | Chubu University Department of Information Science() |
| Date | 2011-07-25 |
| Paper # | NC2011-28 |
| Volume (vol) | vol.111 |
| Number (no) | 157 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |