Optimal Spectral Transitions in High-Dimensional Multi-Index Models

Leonardo Defilippis, Yatin Dandi, Pierre Mergny, Florent Krzakala, Bruno Loureiro · 2025

Read Optimal Spectral Transitions in High-Dimensional Multi-Index Models by Leonardo Defilippis, Yatin Dandi, Pierre Mergny, Florent Krzakala, Bruno Loureiro (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-2502-02545.txt.

Description

Leonardo Defilippis, Yatin Dandi, Pierre Mergny, Florent Krzakala, and Bruno Loureiro find the exact sample threshold for weakly rebuilding a hidden signal space, in a paper from two thousand twenty-five. The test bed is a Gaussian multi-index model, a stand-in for network learning depth. Past results covered the single-index case only.

Spectral methods built from a linearized message scheme hit the best threshold, the team shows. Above that line the top direction lines up with the hidden space, like a known spike jump in random matrix theory. Math proof and number tests agree. Readers in learning theory get a closed gap on weak learnability limits for multi-index models.

Catalog record

CREATORLeonardo Defilippis, Yatin Dandi, Pierre Mergny, Florent Krzakala, Bruno Loureiro
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE108 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-02545

Browse the full library · About the Tower of Babel