| Presentation | 2023-07-06 Convergence Acceleration of Particle-based Variational Inference by Deep Unfolding Yuya Kawamura, Satoshi Takabe, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Stein Variational Gradient Descent(SVGD) is a prominent particle-based variational inference method used for estimating posterior distributions in probabilistic models . Its remarkable approximation capabilities have attracted considerable attention. In this paper, we propose a trainable algorithm that incorporates a deep learning technique known as deep unfolding into SVGD. This approach enables the learning of internal parameters of SVGD, leading to the acceleration of the convergence speed. To evaluate the proposed two trainable SVGD algorithms, we conducted numerical simulations for three different tasks: sampling from one-dimensional Gaussian mixture, Bayesian logistic regression, and Bayesian neural networks. The results show that our proposed algorithms exhibit faster convergence compared to conventional variants of SVGD. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | particle-based variational inference / deep learning / deep unfolding |
| Paper # | CAS2023-8,VLD2023-8,SIP2023-24,MSS2023-8 |
| Date of Issue | 2023-06-29 (CAS, VLD, SIP, MSS) |
| Conference Information | |
| Committee | MSS / CAS / SIP / VLD |
|---|---|
| Conference Date | 2023/7/6(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | Shingo Yamaguchi(Yamaguchi Univ.) / Yasutoshi Aibara(OmniVision) / Takayuki Nakachi(Ryukyu Univ.) / Shigetoshi Nakatake(Univ. of Kitakyushu) |
| Vice Chair | Toshiyuki Miyamoto(Osaka Inst. of Tech.) / Norihiko Shinomiya(Soka Univ.) / Koichi Ichige(Yokohama National Univ.) / Kiyoshi Nishikawa(okyo Metropolitan Univ.) / Yuichi Sakurai(Hitachi) |
| Secretary | Toshiyuki Miyamoto(Osaka Univ.) / Norihiko Shinomiya(NEC) / Koichi Ichige(Soka Univ.) / Kiyoshi Nishikawa(Renesas Electronics) / Yuichi Sakurai(Chiba Univ.) |
| Assistant | Masato Shirai(Shimane Univ.) / Nao Ito(NIT, Toyama college) / Motoi Yamaguchi(TECHNOPRO) / Shinji Shimoda(Sony Semiconductor Solutions) / Shunsuke Koshita(Hachinohe Inst. of Tech.) / Taichi Yoshida(UEC) / Sayaka Shiota(Tokyo Metropolitan Univ.) / Takuma Nishimoto(Hitachi) |
| Paper Information | |
| Registration To | Technical Committee on Mathematical Systems Science and its Applications / Technical Committee on Circuits and Systems / Technical Committee on Signal Processing / Technical Committee on VLSI Design Technologies |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Convergence Acceleration of Particle-based Variational Inference by Deep Unfolding |
| Sub Title (in English) | |
| Keyword(1) | particle-based variational inference |
| Keyword(2) | deep learning |
| Keyword(3) | deep unfolding |
| 1st Author's Name | Yuya Kawamura |
| 1st Author's Affiliation | Tokyo Institute of Technology(Tokyo Tech) |
| 2nd Author's Name | Satoshi Takabe |
| 2nd Author's Affiliation | Tokyo Institute of Technology(Tokyo Tech) |
| Date | 2023-07-06 |
| Paper # | CAS2023-8,VLD2023-8,SIP2023-24,MSS2023-8 |
| Volume (vol) | vol.123 |
| Number (no) | CAS-97,VLD-98,SIP-99,MSS-100 |
| Page | pp.pp.37-42(CAS), pp.37-42(VLD), pp.37-42(SIP), pp.37-42(MSS), |
| #Pages | 6 |
| Date of Issue | 2023-06-29 (CAS, VLD, SIP, MSS) |