Eur. Phys. J. Spec. Top. (2023) 232: 3645-3660
https://doi.org/10.1140/epjs/s11734-023-00844-2

Regular Article

Selective multiple power iteration: from tensor PCA to gradient-based exploration of landscapes

¹ Université Paris-Saclay, CEA-LIST, 91120, Palaiseau, France
² Université Paris-Saclay, CNRS/IN2P3, IJCLab, 91405, Orsay, France

^c Vincent.rivasseau@ijclab.in2p3.fr

Received: 4 June 2022
Accepted: 14 April 2023
Published online: 1 June 2023

Abstract

We propose Selective Multiple Power Iterations (SMPI), a new algorithm to address the important Tensor PCA problem that consists in recovering a spike ${\mathit{v}}_0^{\otimes k}$ corrupted by a Gaussian noise tensor $\mathsf{Z}\in {({\mathbb{R}}^n)}^{\otimes k}$ such that $\mathsf{T}=\sqrt{n}\beta {\mathit{v}}_0^{\otimes k}+\mathsf{Z}$ where $\beta$ is the signal-to-noise ratio. SMPI consists in generating a polynomial number of random initializations, performing a polynomial number of symmetrized tensor power iterations on each initialization, then selecting the one that maximizes $⟨\mathsf{T},{\mathit{v}}^k⟩$. Various numerical simulations for $k=3$ in the conventionally considered range $n\le 1000$, where existent algorithms exhibit negligible finite size effects, show that the experimental performances of SMPI improve drastically upon existent algorithms and becomes comparable to the theoretical optimal recovery. We show that these unexpected performances are due to a powerful mechanism in which the noise plays a key role for the signal recovery and that takes place at low $\beta$. Furthermore, this mechanism results from five essential features of SMPI that distinguish it from previous algorithms based on power iteration. These remarkable results may have strong impact on both practical and theoretical applications of Tensor PCA. (i) We provide in the supplementary material multiple variants of this algorithm to tackle low-rank CP tensor decomposition. These proposed algorithms also outperforms existent methods even on real data which shows a huge potential impact for practical applications. (ii) We present new theoretical insights on the behavior of SMPI and gradient descent methods for the optimization in high-dimensional non-convex landscapes that are present in various machine learning problems. (iii) We expect that these results may help the discussion concerning the existence of the conjectured statistical-algorithmic gap.