Id: arxiv/2602.09959 Url: https://arxiv.org/abs/2602.09959 Tags: math.ST, cs.LG, stat.ML Score: 0 Title: Statistical-Computational Trade-offs in Learning Multi-Index Models via Harmonic Analysis Domain: math.ST Source: arxiv Authors: Hugo Latourelle-Vigeant, Theodor Misiakiewicz Reviews: 0 Subject: math.ST Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-02-10 Page: https://delphi.science/r/arxiv/2602.09959 We study the problem of learning multi-index models (MIMs), where the label depends on the input $\boldsymbol{x} \in \mathbb{R}^d$ only through an unknown $\mathsf{s}$-dimensional projection $\boldsymbol{W}_*^\mathsf{T} \boldsymbol{x} \in \mathbb{R}^\mathsf{s}$. Exploiting the equivariance of this problem under the orthogonal group $\mathcal{O}_d$, we obtain a sharp harmonic-analytic characterization of the learning complexity for MIMs with spherically symmetric inputs -- which refines and generalizes previous Gaussian-specific analyses. Specifically, we derive statistical and computational…