Equivariance by Local Canonicalization: A Matter of Representation
Gerrit Gerhartz ∗ 1 , Peter Lippmann 1 & Fred A. Hamprecht 1 · 2025
Read Equivariance by Local Canonicalization: A Matter of Representation by Gerrit Gerhartz ∗ 1 , Peter Lippmann 1 & Fred A. Hamprecht 1 (2025) 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-2509-26499.txt.
Description
Gerrit Gerhartz, Peter Lippmann, and Fred Hamprecht moved equivariant neural networks onto cheaper machinery in a 2025 paper. Networks for molecules should respect rotations: turning the input should turn the output the same way. Tensor field networks guarantee that, but their special tensor operations run slowly. Local canonicalization instead rotates each neighborhood into a standard pose first, letting ordinary message passing do the work.
Their framework converts existing tensor field designs into that style while preserving the symmetry guarantee and cutting runtime sharply. A systematic comparison weighs representations by theory cost, measured speed, and accuracy. The released tensor_frames package plugs the approach into a standard PyTorch geometry library, so any message passing model can gain equivariance.
Catalog record
| CREATOR | Gerrit Gerhartz ∗ 1 , Peter Lippmann 1 & Fred A. Hamprecht 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2509-26499