| Presentation | 2021-03-03 Evaluation of Concept Drift Detection by monitoring Maximum Safe Radius Naoto Sato, Hironobu Kuruma, Hideto Ogawa, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In recent years, machine-learned software is widely used in a system. In machine learning, a model is trained by collected data. If features of the data change during system operation and the changed data is input to the trained model, the accucary of the traind model decreases (it is called concept drift). When the concept drift occurs, the mode accuracy can be recovered by retraining or additionnal training with the drifed data. Thus, to make a system resilient, it is important to detect a decrease in the accuracy and handle it apporpriately as soon as possible. However, to evaluate the accuracy, it is necessary to define expected output data to each input data, and it needs a lot of human costs. Therefore, it would be better if a decrease of the accuracy by the conecpt drift could be detected without expected data. We assume that the maximum safe radius is useful for the concept drift detection. In this report, this assumption is experimentally evaluated. The results show that it is possible to detect a decrease of the accuracy by monitoring maximum safe radius. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Machine learning / Concept drift / Maximum safe radius |
| Paper # | SS2020-35 |
| Date of Issue | 2021-02-24 (SS) |
| Conference Information | |
| Committee | SS |
|---|---|
| Conference Date | 2021/3/3(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Online |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Takashi Kobayashi(Tokyo Inst. of Tech.) |
| Vice Chair | Kozo Okano(Shinshu Univ.) |
| Secretary | Kozo Okano(Hiroshima City Univ.) |
| Assistant | Shinpei Ogata(Shinshu Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Software Science |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Evaluation of Concept Drift Detection by monitoring Maximum Safe Radius |
| Sub Title (in English) | |
| Keyword(1) | Machine learning |
| Keyword(2) | Concept drift |
| Keyword(3) | Maximum safe radius |
| 1st Author's Name | Naoto Sato |
| 1st Author's Affiliation | Hitachi, Ltd.(Hitachi) |
| 2nd Author's Name | Hironobu Kuruma |
| 2nd Author's Affiliation | Hitachi, Ltd.(Hitachi) |
| 3rd Author's Name | Hideto Ogawa |
| 3rd Author's Affiliation | Hitachi, Ltd.(Hitachi) |
| Date | 2021-03-03 |
| Paper # | SS2020-35 |
| Volume (vol) | vol.120 |
| Number (no) | SS-407 |
| Page | pp.pp.43-48(SS), |
| #Pages | 6 |
| Date of Issue | 2021-02-24 (SS) |