| Presentation | 2016-11-16 Additive Model Decomposition with Global Sparse Structure for Multi-task Granger Causal Estimation Hitoshi Abe, Jun Sakuma, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Causality estimation is one of the key issues in time-series data analysis. Granger causality is widely known as a formulation to find causality among time-series. Predicting responses from past responses and other features, the feature which is significantly useful for prediction is called Granger cause. Existing Granger causality estimation methods are formulated as the feature selection problem by sparse regularizers. One common problem of existing methods is that it captures only the past responses when they overly effect on the prediction. In this paper, we overcome this problem by multi-task learning. We assume coefficients corresponding to the past responses are greater than those of other features on all of the task. We additively decompose a model into task-common model and task-specific models. The task-common model represents the large effect on the past responses and the task-specific models discover Granger cause that rarely appear. In addition, we propose a global sparse regularizer that makes the integrated model which is sum of additive model sparse. Finally, we demonstrate the effectiveness of our proposed method by experiments. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Granger causality / multi-task learning / sparse regularization / model decomposition |
| Paper # | IBISML2016-56 |
| Date of Issue | 2016-11-09 (IBISML) |
| Conference Information | |
| Committee | IBISML |
|---|---|
| Conference Date | 2016/11/16(3days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Kyoto Univ. |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Information-Based Induction Science Workshop (IBIS2016) |
| Chair | Kenji Fukumizu(ISM) |
| Vice Chair | Masashi Sugiyama(Univ. of Tokyo) / Hisashi Kashima(Kyoto Univ.) |
| Secretary | Masashi Sugiyama(Univ. of Tokyo) / Hisashi Kashima(Nagoya Inst. of Tech.) |
| Assistant | Toshihiro Kamishima(AIST) / Tomoharu Iwata(NTT) |
| Paper Information | |
| Registration To | Technical Committee on Infomation-Based Induction Sciences and Machine Learning |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Additive Model Decomposition with Global Sparse Structure for Multi-task Granger Causal Estimation |
| Sub Title (in English) | |
| Keyword(1) | Granger causality |
| Keyword(2) | multi-task learning |
| Keyword(3) | sparse regularization |
| Keyword(4) | model decomposition |
| 1st Author's Name | Hitoshi Abe |
| 1st Author's Affiliation | University of Tsukuba(Univ. Tsukuba) |
| 2nd Author's Name | Jun Sakuma |
| 2nd Author's Affiliation | University of Tsukuba/JST CREST(Univ. Tsukuba) |
| Date | 2016-11-16 |
| Paper # | IBISML2016-56 |
| Volume (vol) | vol.116 |
| Number (no) | IBISML-300 |
| Page | pp.pp.73-79(IBISML), |
| #Pages | 7 |
| Date of Issue | 2016-11-09 (IBISML) |