Topological Exploration of High-Dimensional Empirical Risk Landscapes: general approach, and applications to phase retrieval
Antoine Maillard, Tony Bonnaire, Giulio Biroli · 2026
Read Topological Exploration of High-Dimensional Empirical Risk Landscapes: general approach, and applications to phase retrieval by Antoine Maillard, Tony Bonnaire, Giulio Biroli (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-2602-17779.txt.
Description
Antoine Maillard and team explore the topology of high-dimensional empirical risk landscapes, with an application to phase retrieval. They study risk minimization for high-dimensional Gaussian single-index models, where an unknown signal must be recovered from a loss function built on label pairs. The regime with sample count comparable to dimension is the focus.
Using the Kac-Rice formula, they analyze landscape complexity, defined as the expected number of critical points, across types including local minima. Variational formulas from the literature simplify drastically to explicit problems over finitely many scalar parameters, solvable numerically. The frame also predicts critical point traits, including Hessian spectra and label spreads. Applied to real phase retrieval, it yields full topological phase diagrams, including BBP-type transitions where the Hessian at local minima turns unstable toward the signal. Tests against gradient flow dynamics agree excellently.
Catalog record
| CREATOR | Antoine Maillard, Tony Bonnaire, Giulio Biroli |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 127 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17779