Score-based Data Assimilation

Authors: François Rozet, Gilles Louppe

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present theoretical and empirical evidence supporting the effectiveness of our method. We demonstrate the effectiveness of score-based data assimilation on two chaotic dynamical systems: the Lorenz 1963 [47] and Kolmogorov flow [48] systems.
Researcher Affiliation Academia François Rozet University of Liège francois.rozet@uliege.be Gilles Louppe University of Liège g.louppe@uliege.be
Pseudocode Yes Algorithm 1 Training ϵϕ(xi k:i+k(t), t)
Open Source Code Yes The code for all experiments is made available at https://github.com/francois-rozet/sda.
Open Datasets No We generate 1024 independent trajectories of 1024 states, which are split into training (80 %), validation (10 %) and evaluation (10 %) sets.
Dataset Splits Yes We generate 1024 independent trajectories of 1024 states, which are split into training (80 %), validation (10 %) and evaluation (10 %) sets.
Hardware Specification No Experiments were conducted with the help of a cluster of GPUs. In particular, score networks were trained and evaluated concurrently, each on a single GPU with at least 11 GB of memory.
Software Dependencies No No specific version numbers are provided for software dependencies like jax-cfd or the optimizers and activation functions used.
Experiment Setup Yes Architecture and training details for each k are provided in Appendix D. (e.g., from Table 1: Weight decay 10^-3, Learning rate 10^-3, Batch size 256)