HamiFormer: Dual-Expert Diffusion Fields with Affine Symplectic Maps
HamiFormer cuts Hamiltonian phase-space error using dual-expert diffusion and affine symplectic maps.
HamiFormer is a dual-expert diffusion field that combines whole-window denoising with residual-corrected Hamiltonian propagation and Parallel Local Affine Scan using affine symplectic maps. A Regime Model Tree routes residuals to balance typical accuracy against large tail errors. In 192-step evaluations it cut normalized phase-space MSE by 26.3% versus PhysiFormer on HamiBalls-1 and 21.4% versus DiT on HamiBalls-2, with the lowest late-horizon position and momentum errors among reported baselines.
- Dual-expert diffusion mixes denoising with Hamiltonian residual correction
- Parallel Local Affine Scan uses affine symplectic maps for faster refinement
- Phase-space MSE down 26.3% versus PhysiFormer on HamiBalls-1
- MSE down 21.4% versus DiT on HamiBalls-2 over 192 steps
- AI models
- HamiFormerPhysiFormerDiT
From the original article
1 min read at arxiv.orgPredicting smooth dynamics and collisions requires modeling continuous evolution and abrupt state changes. We introduce HamiFormer, a dual-expert diffusion field combining whole-window denoising with residual-corrected Hamiltonian propagation.
By Haoxiang Huang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.