WEKO3
アイテム
Asynchronous Decentralized Distributed K-FAC : Enhancing Training Efficiency and Load Balancing in Heterogeneous Environments (unreferred)
https://ipsj.ixsq.nii.ac.jp/records/241670
https://ipsj.ixsq.nii.ac.jp/records/241670e68b64ec-3295-4887-9c15-6a5a8be8effb
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
IPSJ-ARC24259005.pdf (1.8 MB)
2026年12月9日からダウンロード可能です。
|
Copyright (c) 2024 by the Information Processing Society of Japan
|
|
| 非会員：¥660, IPSJ:学会員：¥330, ARC:会員：¥0, DLIB:会員：¥0 | ||
| Item type | SIG Technical Reports(1) | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 公開日 | 2024-12-09 | |||||||||
| タイトル | ||||||||||
| タイトル | Asynchronous Decentralized Distributed K-FAC : Enhancing Training Efficiency and Load Balancing in Heterogeneous Environments (unreferred) | |||||||||
| タイトル | ||||||||||
| 言語 | en | |||||||||
| タイトル | Asynchronous Decentralized Distributed K-FAC : Enhancing Training Efficiency and Load Balancing in Heterogeneous Environments (unreferred) | |||||||||
| 言語 | ||||||||||
| 言語 | eng | |||||||||
| キーワード | ||||||||||
| 主題Scheme | Other | |||||||||
| 主題 | 並列計算 | |||||||||
| 資源タイプ | ||||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_18gh | |||||||||
| 資源タイプ | technical report | |||||||||
| 著者所属 | ||||||||||
| Graduate School of Science and Technology, University of Tsukuba | ||||||||||
| 著者所属 | ||||||||||
| Center for Computational Sciences, University of Tsukuba | ||||||||||
| 著者所属(英) | ||||||||||
| en | ||||||||||
| Graduate School of Science and Technology, University of Tsukuba | ||||||||||
| 著者所属(英) | ||||||||||
| en | ||||||||||
| Center for Computational Sciences, University of Tsukuba | ||||||||||
| 著者名 |
Mingzhe, Yu
× Mingzhe, Yu
× Osamu, Tatebe
|
|||||||||
| 著者名(英) |
Mingzhe, Yu
× Mingzhe, Yu
× Osamu, Tatebe
|
|||||||||
| 論文抄録 | ||||||||||
| 内容記述タイプ | Other | |||||||||
| 内容記述 | We propose AD-KFAC, an asynchronous decentralized framework extending the K-FAC optimizer for large-scale distributed deep learning in heterogeneous and unstable environments. Our method incorporates dynamic load balancing and fault tolerance mechanisms to enhance scalability and robustness. By combining an asynchronous K-FAC computation module, an RPC communication module, and dynamic task allocation based on the Raft consensus algorithm, our framework mitigates the impact of straggler nodes and network instability. Experimental results on a 16-node cluster using ResNet-18 on the CIFAR-10 dataset demonstrate that AD-KFAC outperforms the baseline integrated with PyTorch Distributed Data Parallel under varying computation and communication delays, as well as in the presence of node crashes or disconnections. Our method maintains stable convergence and achieves comparable or better final accuracy, showcasing its effectiveness. | |||||||||
| 論文抄録(英) | ||||||||||
| 内容記述タイプ | Other | |||||||||
| 内容記述 | We propose AD-KFAC, an asynchronous decentralized framework extending the K-FAC optimizer for large-scale distributed deep learning in heterogeneous and unstable environments. Our method incorporates dynamic load balancing and fault tolerance mechanisms to enhance scalability and robustness. By combining an asynchronous K-FAC computation module, an RPC communication module, and dynamic task allocation based on the Raft consensus algorithm, our framework mitigates the impact of straggler nodes and network instability. Experimental results on a 16-node cluster using ResNet-18 on the CIFAR-10 dataset demonstrate that AD-KFAC outperforms the baseline integrated with PyTorch Distributed Data Parallel under varying computation and communication delays, as well as in the presence of node crashes or disconnections. Our method maintains stable convergence and achieves comparable or better final accuracy, showcasing its effectiveness. | |||||||||
| 書誌レコードID | ||||||||||
| 収録物識別子タイプ | NCID | |||||||||
| 収録物識別子 | AN10096105 | |||||||||
| 書誌情報 |
研究報告システム・アーキテクチャ（ARC） 巻 2024-ARC-259, 号 5, p. 1-10, 発行日 2024-12-09 |
|||||||||
| ISSN | ||||||||||
| 収録物識別子タイプ | ISSN | |||||||||
| 収録物識別子 | 2188-8574 | |||||||||
| Notice | ||||||||||
| SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. | ||||||||||
| 出版者 | ||||||||||
| 言語 | ja | |||||||||
| 出版者 | 情報処理学会 | |||||||||
