Meta-Learning Dynamics Forecasting Using Task Inference

Authors: Rui Wang, Robin Walters, Rose Yu

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, we demonstrate that our model outperforms state-of-the-art approaches to forecasting complex physical dynamics including turbulent flow, real-world sea surface temperature and ocean currents.
Researcher Affiliation Academia Rui Wang* UC San Diego Robin Walters* Northeastern University Rose Yu UC San Diego
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is available open sourced at https://github.com/Rose-STL-Lab/Dynamic-Adaptation-Network.
Open Datasets Yes We generate a synthetic dataset of turbulent flows with a numerical simulator, Phi Flow1. 1https://github.com/tum-pbs/Phi Flow. We evaluate on a real-world sea surface temperature data generated by the NEMO ocean engine [38]2. 2The data are available at https://resources.marine.copernicus.eu/?option=com_csw&view= details&product_id=GLOBAL_ANALYSIS_FORECAST_PHY_001_024
Dataset Splits Yes For test-future, we train and test on the same task but different time steps. For test-domain, we train and test on different tasks with an 80-20 split.
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments.
Software Dependencies No The paper mentions software like 'Phi Flow' and 'NEMO ocean engine' but does not specify their version numbers or other software dependencies with versions.
Experiment Setup No The paper describes general training strategies like 'multi-step ahead predictions' and 'averaged over 3 runs with random initialization', but does not provide specific hyperparameters such as learning rates, batch sizes, or exact epoch counts for the DyAd model itself in the main text.