| 講演抄録/キーワード |
| 講演名 |
2014-11-18 15:00
[ポスター講演]Tensor Regression and Classification with Latent and Scaled Latent Norms ○Kishan Wimalawarne(Tokyo Inst. of Tech.)・Masashi Sugiyama(Univ. of Tokyo)・Ryota Tomioka(TTIC) IBISML2014-75 |
| 抄録 |
(和) |
(事前公開アブストラクト) In this paper we study tensor based regression and classification us-
ing regularisations with tensor norms. We apply the overlapped norm,
latent norm and the scaled latent norm regularisation to tensor based
learning and give theoretical and experimental evaluations. We derive computationally efficient dual optimisation methods using the alternating direction method of multipliers. We give excess risk bounds for tensor based learning models with different regularisations. Using simulation and real data we demonstrate performances of tensor based regression and classification based on different tensor norms and compare them to vector based learning methods. |
| (英) |
In this paper we study tensor based regression and classification using regularisations with tensor norms. We apply the latent norm and the scaled latent norm regularisation to tensor based learning and give theoretical and experimental evaluations. To solve tensor based regression and classification problems we derive computationally efficient dual optimisation methods using the alternating direction method of multipliers. We give excess risk bounds for tensor based learning models with different regularisations. Using simulated and real data we experimentally demonstrate that the proposed tensor based regression and classification method compares favourably to vector based learning methods. |
| キーワード |
(和) |
/ / / / / / / |
| (英) |
Scaled latent tensor norm / regression / classification / excess loss / / / / |
| 文献情報 |
信学技報, vol. 114, no. 306, IBISML2014-75, pp. 299-306, 2014年11月. |
| 資料番号 |
IBISML2014-75 |
| 発行日 |
2014-11-10 (IBISML) |
| ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
IBISML2014-75 |