MSINO: Curvature-Aware Sobolev Optimization for Manifold Neural Networks
Suresan Pareth · 2026
Read MSINO: Curvature-Aware Sobolev Optimization for Manifold Neural Networks by Suresan Pareth (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-2602-22937.txt.
Description
Suresan Pareth introduces bend-aware Sobolev tuning for neural networks on curved spaces. Standard training checks flat-space slopes, but networks living on curved spaces need geometry-aware learning. MSINO swaps in a matching Sobolev loss that aligns slopes through shifted carry, plus a Laplace smoothness term for steadiness.
Curved-space tuning theory yields shape-dependent constants behind three promises. First, a descent lemma with a curved-space smoothness constant. Second, a Sobolev Polyak-Lojasiewicz rule giving straight-line landing for curved-space gradient descent under plain step bounds. Third, a two-step Newton-Sobolev method with local squared shrink in bend-ruled zones. Unlike flat-space Sobolev training, MSINO gives promises that plainly track bend and carried Jacobians. Uses span surface imaging, physics-informed learning, and robotics on Lie groups. The setup joins value- and slope-based learning with bend-aware landing promises.
Catalog record
| CREATOR | Suresan Pareth |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-22937