English

Near-Optimal Tensor PCA via Normalized Stochastic Gradient Ascent with Overparameterization

Optimization and Control 2025-10-17 v1

Abstract

We study the Order-kk (k4k \geq 4) spiked tensor model for the tensor principal component analysis (PCA) problem: given NN i.i.d. observations of a kk-th order tensor generated from the model T=λvk+E\mathbf{T} = \lambda \cdot v_*^{\otimes k} + \mathbf{E}, where λ>0\lambda > 0 is the signal-to-noise ratio (SNR), vv_* is a unit vector, and E\mathbf{E} is a random noise tensor, the goal is to recover the planted vector vv_*. We propose a normalized stochastic gradient ascent (NSGA) method with overparameterization for solving the tensor PCA problem. Without any global (or spectral) initialization step, the proposed algorithm successfully recovers the signal vv_* when Nλ2Ω~(dk/2)N\lambda^2 \geq \widetilde{\Omega}(d^{\lceil k/2 \rceil}), thereby breaking the previous conjecture that (stochastic) gradient methods require at least Ω(dk1)\Omega(d^{k-1}) samples for recovery. For even kk, the Ω~(dk/2)\widetilde{\Omega}(d^{k/2}) threshold coincides with the optimal threshold under computational constraints, attained by sum-of-squares relaxations and related algorithms. Theoretical analysis demonstrates that the overparameterized stochastic gradient method not only establishes a significant initial optimization advantage during the early learning phase but also achieves strong generalization guarantees. This work provides the first evidence that overparameterization improves statistical performance relative to exact parameterization that is solved via standard continuous optimization.

Keywords

Cite

@article{arxiv.2510.14329,
  title  = {Near-Optimal Tensor PCA via Normalized Stochastic Gradient Ascent with Overparameterization},
  author = {Shihong Ding and Yihong Gu and Yuanshi Liu and Cong Fang},
  journal= {arXiv preprint arXiv:2510.14329},
  year   = {2025}
}
R2 v1 2026-07-01T06:40:31.635Z