Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

FuXi-Ocean: A Global Ocean Forecasting System with Sub-Daily Resolution

Authors: Qiusheng Huang, Yuan Niu, Xiaohui Zhong, AnboyuGuo, Lei Chen, dianjun zhang, Xuefeng Zhang, Hao Li

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through comprehensive experimental evaluation, Fu Xi-Ocean demonstrates superior skill in predicting key variables, including temperature, salinity, and currents, across multiple depths. (...) 4 Experiment (...) We train and evaluate Fu Xi-Ocean using the HYCOM Reanalysis Data [8], the only publicly available ocean dataset with 6-hour temporal resolution.
Researcher Affiliation Academia 1Artificial Intelligence Innovation and Incubation Institute, Fudan University 2Shanghai Innovation Institute 3School of Marine Science and Technology,Tianjin University
Pseudocode No The paper describes the model architecture with diagrams (Figure 1) and textual descriptions in Section 3.2, but it does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No Code, data, and checkpoints will be released publicly upon acceptance.
Open Datasets Yes We train and evaluate Fu Xi-Ocean using the HYCOM Reanalysis Data [8], the only publicly available ocean dataset with 6-hour temporal resolution. (...) For evaluation against real-world observations, we employ the GODAE Ocean View Intercomparison and Validation Task Team (IV-TT) Class 4 framework [39], which is obtained from publicly available sources.
Dataset Splits Yes We train and evaluate Fu Xi-Ocean using the HYCOM Reanalysis Data [8]... We selecte approximately 8.5 years (January 2006 to June 2014) of data for training, a six-month period (July to December 2014) for validation, and one year (January 2015 to December 2015) for testing, focusing on 20 strategically chosen vertical layers down to 1500 m that capture essential ocean dynamics.
Hardware Specification Yes We train on a cluster of 4 NVIDIA H100 GPUs for 60,000 iterations with a batch size of 1 per GPU, requiring approximately 81 hours to complete.
Software Dependencies No The Fu Xi-Ocean employs the Py Torch framework [35] with the Adam W [23, 29] optimizer, configured with β1 = 0.9, β2 = 0.95, and a cosine annealing learning rate schedule [30] that decays from 2.5 10 4 to 10 8.
Experiment Setup Yes The Fu Xi-Ocean employs the Py Torch framework [35] with the Adam W [23, 29] optimizer, configured with β1 = 0.9, β2 = 0.95, and a cosine annealing learning rate schedule [30] that decays from 2.5 10 4 to 10 8. We train on a cluster of 4 NVIDIA H100 GPUs for 60,000 iterations with a batch size of 1 per GPU, requiring approximately 81 hours to complete.