Intelligent Sampling of Extreme-Scale Turbulence Datasets for Accurate and Efficient Spatiotemporal Model Training

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Brewer, Wesley, Gopalakrishnan, Murali Meena, Maiterth, Matthias, Kashi, Aditya, Choi, Jong Youl, Zhang, Pei, Nichols, Stephen, Balin, Riccardo, Couchman, Miles, Kops, Stephen de Bruyn, Yeung, P. K., Dotson, Daniel, Uma-Vaideswaran, Rohini, Oral, Sarp, Wang, Feiyi
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915573284208640
author Brewer, Wesley
Gopalakrishnan, Murali Meena
Maiterth, Matthias
Kashi, Aditya
Choi, Jong Youl
Zhang, Pei
Nichols, Stephen
Balin, Riccardo
Couchman, Miles
Kops, Stephen de Bruyn
Yeung, P. K.
Dotson, Daniel
Uma-Vaideswaran, Rohini
Oral, Sarp
Wang, Feiyi
author_facet Brewer, Wesley
Gopalakrishnan, Murali Meena
Maiterth, Matthias
Kashi, Aditya
Choi, Jong Youl
Zhang, Pei
Nichols, Stephen
Balin, Riccardo
Couchman, Miles
Kops, Stephen de Bruyn
Yeung, P. K.
Dotson, Daniel
Uma-Vaideswaran, Rohini
Oral, Sarp
Wang, Feiyi
contents With the end of Moore's law and Dennard scaling, efficient training increasingly requires rethinking data volume. Can we train better models with significantly less data via intelligent subsampling? To explore this, we develop SICKLE, a sparse intelligent curation framework for efficient learning, featuring a novel maximum entropy (MaxEnt) sampling approach, scalable training, and energy benchmarking. We compare MaxEnt with random and phase-space sampling on large direct numerical simulation (DNS) datasets of turbulence. Evaluating SICKLE at scale on Frontier, we show that subsampling as a preprocessing step can, in many cases, improve model accuracy and substantially lower energy consumption, with observed reductions of up to 38x.
format Preprint
id arxiv_https___arxiv_org_abs_2508_03872
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Intelligent Sampling of Extreme-Scale Turbulence Datasets for Accurate and Efficient Spatiotemporal Model Training
Brewer, Wesley
Gopalakrishnan, Murali Meena
Maiterth, Matthias
Kashi, Aditya
Choi, Jong Youl
Zhang, Pei
Nichols, Stephen
Balin, Riccardo
Couchman, Miles
Kops, Stephen de Bruyn
Yeung, P. K.
Dotson, Daniel
Uma-Vaideswaran, Rohini
Oral, Sarp
Wang, Feiyi
Machine Learning
Artificial Intelligence
Distributed, Parallel, and Cluster Computing
With the end of Moore's law and Dennard scaling, efficient training increasingly requires rethinking data volume. Can we train better models with significantly less data via intelligent subsampling? To explore this, we develop SICKLE, a sparse intelligent curation framework for efficient learning, featuring a novel maximum entropy (MaxEnt) sampling approach, scalable training, and energy benchmarking. We compare MaxEnt with random and phase-space sampling on large direct numerical simulation (DNS) datasets of turbulence. Evaluating SICKLE at scale on Frontier, we show that subsampling as a preprocessing step can, in many cases, improve model accuracy and substantially lower energy consumption, with observed reductions of up to 38x.
title Intelligent Sampling of Extreme-Scale Turbulence Datasets for Accurate and Efficient Spatiotemporal Model Training
topic Machine Learning
Artificial Intelligence
Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2508.03872