Presentation | 2021-12-17 Data Selection for Efficient Deep Learning Ryota Higashi, Toshikazu Wada, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We are investigating the method to sample the important data from the whole dataset for efficient training of Deep Neural Networks. In this report, we compare the accuracy of image classifiers trained from reduced datasets: sampled near the decision boundaries, sampled uniformly over the latent space, and sampled randomly. Experimental results imply that the optimal sampling varies depending on the number of samples, and the data selection criteria should be changed accordingly. Also, we introduce distillation to recover the accuracy degradation by the data reduction and evaluate its effect. Data selection can be applied to many problems: prototyping, active learning, and weighted learning. Findings in this report can be utilized to produce better solution of them. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Deep Learning / Data Selection / Image Classification / Distillation / Active Learning |
Paper # | PRMU2021-51 |
Date of Issue | 2021-12-09 (PRMU) |
Conference Information | |
Committee | PRMU |
---|---|
Conference Date | 2021/12/16(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Online |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Seiichi Uchida(Kyushu Univ.) |
Vice Chair | Masakazu Iwamura(Osaka Pref. Univ.) / Mitsuru Anpai(Denso IT Lab.) |
Secretary | Masakazu Iwamura(NTT) / Mitsuru Anpai(Tottori Univ.) |
Assistant | Kouta Yamaguchi(CyberAgent) / Yusuke Matsui(Univ. of Tokyo) |
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) | Data Selection for Efficient Deep Learning |
Sub Title (in English) | |
Keyword(1) | Deep Learning |
Keyword(2) | Data Selection |
Keyword(3) | Image Classification |
Keyword(4) | Distillation |
Keyword(5) | Active Learning |
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 | 2021-12-17 |
Paper # | PRMU2021-51 |
Volume (vol) | vol.121 |
Number (no) | PRMU-304 |
Page | pp.pp.148-153(PRMU), |
#Pages | 6 |
Date of Issue | 2021-12-09 (PRMU) |