Presentation | 2020-12-18 Regularization Using Knowledge Distillation in Learning Small Datasets Ryota Higashi, Toshikazu Wada, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | Knowledge distillation is a method mainly used for compressing deep learning models, but it has recently gained attention for its effectiveness in learning from small amounts of data as well. In this report, taking the image classification problem as an example, we focused on the fact that the decrease in classification accuracy can be suppressed by distillation when the training data is reduced, and the accuracy varies with a distillation parameter called “temperature”. First, we prepare a teacher model trained on all the training data and then distill it into a student model. In this case, we found that the accuracy of the student model is improved by increasing the temperature, especially when the number of training data is small, and that this effect is not related to the calibration of the teacher model. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Deep Learning / Knowledge Distillation / Image Classification / Few-Shot Learning / Calibration |
Paper # | PRMU2020-61 |
Date of Issue | 2020-12-10 (PRMU) |
Conference Information | |
Committee | PRMU |
---|---|
Conference Date | 2020/12/17(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Online |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Transfer learning and few shot learning |
Chair | Yoichi Sato(Univ. of Tokyo) |
Vice Chair | Akisato Kimura(NTT) / Masakazu Iwamura(Osaka Pref. Univ.) |
Secretary | Akisato Kimura(Mobility Technologies) / Masakazu Iwamura(Chubu Univ.) |
Assistant | Takashi Shibata(NTT) / Masashi Nishiyama(Tottori Univ.) |
Paper Information | |
Registration To | Technical Committee on Pattern Recognition and Media Understanding |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Regularization Using Knowledge Distillation in Learning Small Datasets |
Sub Title (in English) | |
Keyword(1) | Deep Learning |
Keyword(2) | Knowledge Distillation |
Keyword(3) | Image Classification |
Keyword(4) | Few-Shot Learning |
Keyword(5) | Calibration |
1st Author's Name | Ryota Higashi |
1st Author's Affiliation | Wakayama University(Wakayama Univ.) |
2nd Author's Name | Toshikazu Wada |
2nd Author's Affiliation | Wakayama University(Wakayama Univ.) |
Date | 2020-12-18 |
Paper # | PRMU2020-61 |
Volume (vol) | vol.120 |
Number (no) | PRMU-300 |
Page | pp.pp.133-138(PRMU), |
#Pages | 6 |
Date of Issue | 2020-12-10 (PRMU) |