Learning Tangent Bundles and Characteristic Classes with Autoencoder Atlases
Eduardo Paluzo-Hidalgo, Yuichi Ike · 2026
Read Learning Tangent Bundles and Characteristic Classes with Autoencoder Atlases by Eduardo Paluzo-Hidalgo, Yuichi Ike (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-2602-22873.txt.
Description
Eduardo Paluzo-Hidalgo and one co-author connect multi-chart auto-coders with vector bundles and characteristic classes. Rather than forcing one global flat placing, they treat locally trained encoder-decoder pairs as a learned atlas on a shape, a standard shape viewpoint applied to machine learning.
Any reconstruction-consistent atlas canonically defines transition maps obeying the cocycle condition. Linearizing those maps yields a vector bundle that matches the tangent bundle when latent dimension equals intrinsic dimension. That construction unlocks differential-topological invariants: the first Stiefel-Whitney class falls out of Jacobian signs of learned transition maps, giving an algorithmic orientability test. Nontrivial classes also block single-chart representations, and the minimum chart count follows the manifold's good cover structure.
Catalog record
| CREATOR | Eduardo Paluzo-Hidalgo, Yuichi Ike |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 128 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-22873