Balancing Symmetry and Efficiency in Graph Flow Matching
Benjamin Honoré, Alba Carballo-Castro, Yiming Qin, Pascal Frossard · 2026
Read Balancing Symmetry and Efficiency in Graph Flow Matching by Benjamin Honoré, Alba Carballo-Castro, Yiming Qin, Pascal Frossard (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-18084.txt.
Description
Benjamin Honoré and team study the trade-off between symmetry and speed in graph flow matching. Equivariance is central to graph making models because it respects the swap balance of graphs. Strict versions, however, raise run cost through added design constraints and slow learning, since the model must stay steady across a large space of possible node swaps.
Starting from an equivariant discrete flow-matching model, the authors relax it during training through a tunable symmetry tuning scheme based on sine place codes and node swaps. Tests show breaking symmetry speeds early training with an easier learning signal, but spawns shortcut answers that overfit, with the model often making copies of the training set. Tuning the symmetry signal well delays overfit while speeding learning, reaching stronger results with about a fifth of the standard training epochs.
Catalog record
| CREATOR | Benjamin Honoré, Alba Carballo-Castro, Yiming Qin, Pascal Frossard |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-18084