The authors treated simulation fidelity as a fine-grained discrete spectrum, rather than a binary high/low choice, and trained neural surrogates on mixtures of samples generated at different numerical fidelities.
Mixing simulation fidelities lowers data generation costs for neural PDE surrogates
Training on a tuned mixture of low- and high-fidelity simulations matches high-fidelity-only accuracy while reducing simulator compute by up to 6.3x across six PDE benchmarks.
AI Startup
Paul Setinek · Fabian Paischer · Nils Thuerey · Johannes Brandstetter · Felix Koehler
JKU Linz · Mistral AI · Technical University of Munich · Munich Center for Machine Learning
Research Digest··2 min read
Setinek et al.
Why this paper
From Mistral AI and 3 others
In one line
Training neural PDE surrogates on an optimal mixture of simulation fidelities matches high-fidelity-only accuracy while using up to 6.3x less data generation compute.
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 (3 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.
§