Zobrazeno 1 - 10
of 26
pro vyhledávání: '"Yoon, Daegun"'
Autor:
Yoon, Daegun, Oh, Sangyoon
Communication overhead is a major obstacle to scaling distributed training systems. Gradient sparsification is a potential optimization approach to reduce the communication volume without significant loss of model fidelity. However, existing gradient
Externí odkaz:
http://arxiv.org/abs/2402.13781
Autor:
Yoon, Daegun, Oh, Sangyoon
Gradient sparsification is a communication optimisation technique for scaling and accelerating distributed deep neural network (DNN) training. It reduces the increasing communication traffic for gradient aggregation. However, existing sparsifiers hav
Externí odkaz:
http://arxiv.org/abs/2310.00967
Autor:
Yoon, Daegun, Oh, Sangyoon
Gradient sparsification is a widely adopted solution for reducing the excessive communication traffic in distributed deep learning. However, most existing gradient sparsifiers have relatively poor scalability because of considerable computational cos
Externí odkaz:
http://arxiv.org/abs/2307.03500
Autor:
Yoon, Daegun, Oh, Sangyoon
To train deep learning models faster, distributed training on multiple GPUs is the very popular scheme in recent years. However, the communication bandwidth is still a major bottleneck of training performance. To improve overall training performance,
Externí odkaz:
http://arxiv.org/abs/2209.08497
Publikováno v:
In Journal of Parallel and Distributed Computing December 2022 170:13-23
Publikováno v:
Journal of Supercomputing. Jul2023, Vol. 79 Issue 10, p11387-11409. 23p.
Publikováno v:
Journal of Supercomputing. Apr2023, Vol. 79 Issue 6, p6889-6917. 29p.
Autor:
Yoon, Daegun1 (AUTHOR) kljp@ajou.ac.kr, Oh, Sangyoon1 (AUTHOR) syoh@ajou.ac.kr
Publikováno v:
Sensors (14248220). Jul2022, Vol. 22 Issue 13, p4899-N.PAG. 17p.
Publikováno v:
Journal of Supercomputing. Dec2021, Vol. 77 Issue 12, p13676-13702. 27p.
Akademický článek
Tento výsledek nelze pro nepřihlášené uživatele zobrazit.
K zobrazení výsledku je třeba se přihlásit.
K zobrazení výsledku je třeba se přihlásit.