| Presentation | 2022-06-27 A Bagging Method to Improve the Accuracy of Gaussian Process Regression for Neural Architecture Search Rion Hada, Masao Okita, Fumihiko Ino, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | The goal of this study is to improve performance estimation for neural network architectures in neural architecture search (NAS), which leverages Bayesian optimization with Gaussian process regression. To achieve this goal, we propose a bagging method to boost the accuracy of Gaussian process regression by controlling over-fitting. Aiming to reduce the estimation error with Gaussian process regression, the proposed method extends the acquire function for Bayesian optimization in an existing NAS method: the extended acquire function iteratively estimates the inference accuracy of the target architecture with different supervised datasets and returns the median accuracy. For the rest, as with the existing method, the proposed method searches neural architectures efficiently by limiting the architectures to be actually trained only to those estimated to show high inference accuracy by Bayesian optimization. Experimental results show that the proposed method increased Spearman's rank correlation coefficient between an estimated ranking and the true ranking of inference accuracy for 100 neural architectures from 0.772 to 0.829. This indicates that the proposed method is useful for precisely estimating the inference accuracy of neural architectures. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | NAS (neural architecture search) / surrogate model / Bayesian optimization / Gaussian process regression / overfitting / bagging (bootstrap aggregating) |
| Paper # | NC2022-2,IBISML2022-2 |
| Date of Issue | 2022-06-20 (NC, IBISML) |
| Conference Information | |
| Committee | NC / IBISML / IPSJ-BIO / IPSJ-MPS |
|---|---|
| Conference Date | 2022/6/27(3days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Hiroshi Yamakawa(Univ of Tokyo) / Masashi Sugiyama(Univ. of Tokyo) |
| Vice Chair | Hirokazu Tanaka(Tokyo City Univ.) / Toshihiro Kamishima(AIST) / Koji Tsuda(Univ. of Tokyo) |
| Secretary | Hirokazu Tanaka(NTT) / Toshihiro Kamishima(NICT) / Koji Tsuda(NTT) / (Hokkaido Univ.) |
| Assistant | Yoshimasa Tawatsuji(Waseda Univ.) / Tomoki Kurikawa(KMU) / Yoshinobu Kawahara(Osaka Univ.) / Taiji Suzuki(Tokyo Inst. of Tech.) |
| Paper Information | |
| Registration To | Technical Committee on Neurocomputing / Technical Committee on Infomation-Based Induction Sciences and Machine Learning / Special Interest Group on Bioinformatics and Genomics / Special Interest Group on Mathematical Modeling and Problem Solving |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A Bagging Method to Improve the Accuracy of Gaussian Process Regression for Neural Architecture Search |
| Sub Title (in English) | |
| Keyword(1) | NAS (neural architecture search) |
| Keyword(2) | surrogate model |
| Keyword(3) | Bayesian optimization |
| Keyword(4) | Gaussian process regression |
| Keyword(5) | overfitting |
| Keyword(6) | bagging (bootstrap aggregating) |
| 1st Author's Name | Rion Hada |
| 1st Author's Affiliation | Osaka University(Osaka Univ.) |
| 2nd Author's Name | Masao Okita |
| 2nd Author's Affiliation | Osaka University(Osaka Univ.) |
| 3rd Author's Name | Fumihiko Ino |
| 3rd Author's Affiliation | Osaka University(Osaka Univ.) |
| Date | 2022-06-27 |
| Paper # | NC2022-2,IBISML2022-2 |
| Volume (vol) | vol.122 |
| Number (no) | NC-89,IBISML-90 |
| Page | pp.pp.6-13(NC), pp.6-13(IBISML), |
| #Pages | 8 |
| Date of Issue | 2022-06-20 (NC, IBISML) |