Structured Approximation of Toeplitz Matrices and Subspaces
By: Albert Fannjiang, Weilin Li
Potential Business Impact:
Fixes broken data using math tricks.
This paper studies two structured approximation problems: (1) Recovering a corrupted low-rank Toeplitz matrix and (2) recovering the range of a Fourier matrix from a single observation. Both problems are computationally challenging because the structural constraints are difficult to enforce directly. We show that both tasks can be solved efficiently and optimally by applying the Gradient-MUSIC algorithm for spectral estimation. For a rank $r$ Toeplitz matrix ${\boldsymbol T}\in {\mathbb C}^{n\times n}$ that satisfies a regularity assumption and is corrupted by an arbitrary ${\boldsymbol E}\in {\mathbb C}^{n\times n}$ such that $\|{\boldsymbol E}\|_2\leq αn$, our algorithm outputs a Toeplitz matrix $\widehat{\boldsymbol T}$ of rank exactly $r$ such that $\|{\boldsymbol T}-\widehat{\boldsymbol T}\|_2 \leq C \sqrt r \, \|{\boldsymbol E}\|_2$, where $C,α>0$ are absolute constants. This performance guarantee is minimax optimal in $n$ and $\|{\boldsymbol E}\|_2$. We derive optimal results for the second problem as well. Our analysis provides quantitative connections between these two problems and spectral estimation. Our results are equally applicable to Hankel matrices with superficial modifications.
Similar Papers
Spectral Perturbation Bounds for Low-Rank Approximation with Applications to Privacy
Machine Learning (CS)
Protects private data while finding important patterns.
Sublinear Time Low-Rank Approximation of Hankel Matrices
Data Structures and Algorithms
Finds hidden patterns in data much faster.
Near-optimal Rank Adaptive Inference of High Dimensional Matrices
Information Theory
Finds hidden patterns in messy data.