| Presentation | 2018-11-12 [Invited Talk] Robust Optimization and its Application to Supervised Learning Akiko Takeda, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | There are various uncertainties in real-world problems. When formulating them as mathematical optimization problems, we often use uncertain data such as "data containing measurement error'' or "predicted future demand using past data''. Under such circumstances, a robust optimization method aiming at obtaining robust solutions against data uncertainty is often used. In robust optimization, a range (uncertainty set) where uncertain data will be observed is set in advance, and a mathematical optimization model is constructed so as to find optimal decision under the worst-case scenario in the uncertainty set. The optimal decision by robust optimization does not violate the constraints of the model as long as uncertain data exists in the uncertainty set nor does the objective function value worsen seriously, so a robust solution to uncertainty can be obtained. We will explain the idea and basic formulation of robust optimization and introduce examples of application to supervised machine learning. In the field of machine learning, mathematical optimization methods are often used to find rules and patterns from data. Under the assumption that a given dataset contains errors and is uncertain, some examples of uncertainty sets for uncertain data are shown. Furthermore, we prove that robust optimization models using these uncertainty sets are equivalent to existing supervised machine learning models. In other words, unified interpretation based on robust optimization is possible for supervised machine learning models with completely different input data and formulation, and in addition, it can be seen that the difference between these machine learning models lies in the way of expressing the uncertainty set assumed for uncertain data. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | robust optimization / uncertainty set / machine learning / supervised learning / binary classification |
| Paper # | CAS2018-67,MSS2018-43 |
| Date of Issue | 2018-11-05 (CAS, MSS) |
| Conference Information | |
| Committee | MSS / CAS / IPSJ-AL |
|---|---|
| Conference Date | 2018/11/12(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Morikazu Nakamura(Univ. of Ryukyus) / Hideaki Okazaki(Shonan Inst. of Tech.) |
| Vice Chair | Shigemasa Takai(Osaka Univ.) / Taizo Yamawaki(Hitachi) |
| Secretary | Shigemasa Takai(Toshiba) / Taizo Yamawaki(Osaka Univ.) / (Shonan Inst. of Tech.) |
| Assistant | Hideki Kinjo(Okinawa Univ.) / Motoi Yamaguchi(Renesas Electronics) |
| Paper Information | |
| Registration To | Technical Committee on Mathematical Systems Science and its applications / Technical Committee on Circuits and Systems / Special Interest Group on Algorithms |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | [Invited Talk] Robust Optimization and its Application to Supervised Learning |
| Sub Title (in English) | |
| Keyword(1) | robust optimization |
| Keyword(2) | uncertainty set |
| Keyword(3) | machine learning |
| Keyword(4) | supervised learning |
| Keyword(5) | binary classification |
| 1st Author's Name | Akiko Takeda |
| 1st Author's Affiliation | The University of Tokyo(U.Tokyo) |
| Date | 2018-11-12 |
| Paper # | CAS2018-67,MSS2018-43 |
| Volume (vol) | vol.118 |
| Number (no) | CAS-295,MSS-296 |
| Page | pp.pp.55-55(CAS), pp.55-55(MSS), |
| #Pages | 1 |
| Date of Issue | 2018-11-05 (CAS, MSS) |