FourCastNeXt：面向有限算力优化 FourCastNet 训练
计算机视觉与模式识别
2024-03-22 v2 人工智能
摘要
FourCastNeXt 是对全球机器学习天气预报模型 FourCastNet 的优化，其精度水平相当，且仅需约 5% 的原始 FourCastNet 计算需求即可完成训练。本技术报告介绍了在保持建模变量均方根误差（RMSE）所衡量的相似性能的同时进行模型优化的策略。通过提供训练成本极低的模型，FourCastNeXt 使神经地球系统建模更易于研究人员开展训练实验和消融研究。FourCastNeXt 的训练和推理代码可在 https://github.com/nci/FourCastNeXt 获取。
引用
@article{arxiv.2401.05584,
title = {FourCastNeXt: Optimizing FourCastNet Training for Limited Compute},
author = {Edison Guo and Maruf Ahmed and Yue Sun and Rui Yang and Harrison Cook and Tennessee Leeuwenburg and Ben Evans},
journal= {arXiv preprint arXiv:2401.05584},
year = {2024}
}
备注
Major revision. All prior content (text, figures, table) has been updated. Additionally, new text, tables and figures have been added. Updated title. Updated author list