PDE-OBS integrates numerical simulation, observation construction, model training and evaluation for two tasks: reconstructing a complete stationary field from partial measurements, and forecasting future states from a partially observed initial state.
PDE models falter when sensor patterns change between training and testing
Across seven physical systems and seven methods, models consistently performed worse when evaluated with observation layouts unlike those used for training.
Academic
Ruichen Xu · Siyao Wang · Fang Wan · Jiacheng Qiu · Wenhan Gao · Jiaxing Zhang · +4 more
Stony Brook University · University of California, Davis · Independent Research · PayPal · New York University
Research Digest··3 min read
Thread:PDE Field Reconstruction
Xu and colleagues introduce PDE-OBS, a benchmark for testing how scientific machine-learning models respond to changes in measurement density and spatial layout.
Why this paper
From Stony Brook University and 5 others
In one line
Physical-field models perform worse when test-time measurement layouts differ from training, and additional measurements do not reliably improve a fixed model.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§