MASF: Measurement-Aware Score-based Filter
This repo contains the official implementation of Rethinking Forward Processes for Score-Based Nonlinear Data Assimilation in High Dimensions.
The code and documentation are currently being organized and will be updated progressively.
Contents
2.1. Results at 128 × 128 Resolution
2.2. High-Resolution Results with Grid Mask Measurements
2.3. Temporal RMSE Evolution with Speed Measurement
masf/
├── configs/
├── dynamics/ # Kolmogorov flow
├── measurements/ # Grid mask, Center mask, Sigmoid, Speed
├── methods/ # EnKF, LETKF, SF, SSLS, MASF
├── models/ # UNet, dual-UNet
├── main.py # Run a single method
├── swap/
│ ├── run_tuning.py # Run hyperparameter tuning phases
│ ├── run_post_eval.py # Run post-evaluation and sensitivity analysis
│ ├── configs/
│ └── src/ # Source code for tuning, post evaluation, and reports
├── utils/
├── README.md
└── requirements.txt
git clone git@github.com:tcnllab-oss/masf.git
cd masf
conda create -n masf python=3.10
conda activate masf
pip install -r requirements.txt
python main.py \
--dynamic_type kolmogorov_128 \
--measurement_type grid_mask \
--method_type ours \
--seed 0 \
--exp grid128_masf_seed0
A Colab tutorial will be available soon.
python swap/run_tuning.py \
--until finetuning \
--dynamic_type kolmogorov \
--measurement_type grid_mask \
--method_type ours \
--dim 128
python swap/run_post_eval.py \
--phase num_sample \
--root_dir outputs/kolmogorov_128/grid_mask/ours
@article{yoon2026masf,
title={Rethinking Forward Processes for Score-Based Nonlinear Data Assimilation in High Dimensions},
author={Yoon, Eunbi and Chang, Won and Kim, Donghan and Kim, Dae Wook},
journal={arXiv preprint},
year={2026}
}
For questions, please contact Eunbi Yoon at eunbiyoon6286@kaist.ac.kr.