ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 1804.01221
72
39
v1v2 (latest)

Tight Query Complexity Lower Bounds for PCA via Finite Sample Deformed Wigner Law

4 April 2018
Max Simchowitz
A. Alaoui
Benjamin Recht
ArXiv (abs)PDFHTML
Abstract

We prove a \emph{query complexity} lower bound for approximating the top rrr dimensional eigenspace of a matrix. We consider an oracle model where, given a symmetric matrix M∈Rd×d\mathbf{M} \in \mathbb{R}^{d \times d}M∈Rd×d, an algorithm Alg\mathsf{Alg}Alg is allowed to make T\mathsf{T}T exact queries of the form w(i)=Mv(i)\mathsf{w}^{(i)} = \mathbf{M} \mathsf{v}^{(i)}w(i)=Mv(i) for iii in {1,...,T}\{1,...,\mathsf{T}\}{1,...,T}, where v(i)\mathsf{v}^{(i)}v(i) is drawn from a distribution which depends arbitrarily on the past queries and measurements {v(j),w(i)}1≤j≤i−1\{\mathsf{v}^{(j)},\mathsf{w}^{(i)}\}_{1 \le j \le i-1}{v(j),w(i)}1≤j≤i−1​. We show that for every gap∈(0,1/2]\mathtt{gap} \in (0,1/2]gap∈(0,1/2], there exists a distribution over matrices M\mathbf{M}M for which 1) gapr(M)=Ω(gap)\mathrm{gap}_r(\mathbf{M}) = \Omega(\mathtt{gap})gapr​(M)=Ω(gap) (where gapr(M)\mathrm{gap}_r(\mathbf{M})gapr​(M) is the normalized gap between the rrr and r+1r+1r+1-st largest-magnitude eigenvector of M\mathbf{M}M), and 2) any algorithm Alg\mathsf{Alg}Alg which takes fewer than const×rlog⁡dgap\mathrm{const} \times \frac{r \log d}{\sqrt{\mathtt{gap}}}const×gap​rlogd​ queries fails (with overwhelming probability) to identity a matrix V^∈Rd×r\widehat{\mathsf{V}} \in \mathbb{R}^{d \times r}V∈Rd×r with orthonormal columns for which ⟨V^,MV^⟩≥(1−const×gap)∑i=1rλi(M)\langle \widehat{\mathsf{V}}, \mathbf{M} \widehat{\mathsf{V}}\rangle \ge (1 - \mathrm{const} \times \mathtt{gap})\sum_{i=1}^r \lambda_i(\mathbf{M})⟨V,MV⟩≥(1−const×gap)∑i=1r​λi​(M). Our bound requires only that ddd is a small polynomial in 1/gap1/\mathtt{gap}1/gap and rrr, and matches the upper bounds of Musco and Musco '15. Moreover, it establishes a strict separation between convex optimization and \emph{randomized}, "strict-saddle" non-convex optimization of which PCA is a canonical example: in the former, first-order methods can have dimension-free iteration complexity, whereas in PCA, the iteration complexity of gradient-based methods must necessarily grow with the dimension.

View on arXiv
Comments on this paper