Learning under covariate shift for domain adaptation for word sense disambiguation through weight setting using a outlier detection method
| Project/Area Number |
26330244
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | Ibaraki University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2014-04-01 – 2017-03-31
|
| Project Status |
Completed (Fiscal Year 2016)
|
| Budget Amount *help |
¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000)
Fiscal Year 2016: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000)
Fiscal Year 2015: ¥1,820,000 (Direct Cost: ¥1,400,000、Indirect Cost: ¥420,000)
Fiscal Year 2014: ¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
|
|---|
| Keywords | 語義曖昧性解消 / 領域適応 / 共変量シフト / 外れ値検出 / 重み付き学習 / 確率密度比 |
|---|
| Outline of Final Research Achievements |
In this research, I solved the domain adaptation for word sense disambiguation by using the learning method under the assumption of covariate shift. The key point of this approach is how to estimation of the probability density ratio, and how to conduct the weighted learning. For the first problem, I adopt unconstrained least squares importance fitting (uLSIF). In this research, I showed that a linear kernel is better than a Gaussian kernel used as the basis function generally. Furthermore, I proposed to use 3 kinds of discrete values as a weight. For the second problem, I showed that SVM also is available but the maximum entropy method. Furthermore, I combined the kernel function, the weighted learning and discrete weights.
|
Report
(4 results)
Research Products
(33 results)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
[Presentation] 語義曖昧性解消の誤り分析2015
Author(s)
新納浩幸, 白井清昭, 村田真樹, 福本文代, 藤田早苗, 佐々木稔, 古宮嘉那子, 乾孝司
Organizer
言語処理学会第21回年次大会 ワークショップ「エラー分析ワークショップ」
Place of Presentation
京都大学
Year and Date
2015-03-16 – 2015-03-21
Related Report
-
-
-
-
-
-
-
-
-
-
