The authors trained and tested two neural surrogate architectures: Transolver, a physics-attention transformer, and Bi-Stride Multi-Scale MeshGraphNet (BSMS-MGN), a hierarchical multi-scale graph neural network.
Neural surrogates accelerate linear radiation transport simulations on key benchmarks
Ablation study reveals architecture-dependent inductive bias preferences for two neural surrogate models applied to Lattice and Hohlraum problems.
Big Tech
Carmelo Gonzales · Steffen Schotthöfer · Cory D. Hauck
NVIDIA · Seamless Labs · Oak Ridge National Laboratory
Research Digest··2 min read
The authors benchmarked two neural surrogate architectures, Transolver and BSMS-MGN, as end-to-end approximations of the final-time particle concentration for the two-dimensional linear radiation transport equation on canonical Lattice and Hohlraum benchmarks.
Why this paper
From NVIDIA and 2 others · Released code and data
In one line
On Lattice and Hohlraum radiation-transport benchmarks, neural surrogates usually beat coarse simulations, but the best features and losses depend on architecture and target quantity.
What it released
CodeCodeData
What we could check
- ✓Code link in the paper (github.com)
- ·No weights link found
- ✓Dataset link in the paper (huggingface.co)
- ·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.
§