Detection Thresholds in Very Sparse Matrix Completion

نویسندگان

چکیده

We study the matrix completion problem: an underlying $$m \times n$$ P is low rank, with incoherent singular vectors, and a random A equal to on (uniformly) subset of entries size dn. All other are zero. The goal retrieve information from observation A. Let $$A_1$$ be where each entry multiplied by independent $$\{0,1\}$$ -Bernoulli variable parameter 1/2. This paper about when, how why non-Hermitian eigen-spectra matrices $$A_1 (A - A_1)^*$$ $$(A-A_1)^*A_1$$ captures more relevant principal component structure than $$A A^*$$ $$A^* A$$ . show that eigenvalues asymmetric $$A_{1} A_{1})^{*}$$ $$(A-A_{1})^{*} A_{1}$$ modulus greater detection threshold asymptotically $$PP^*$$ $$P^*P$$ associated eigenvectors aligned as well. central surprise intentionally inducing asymmetry additional randomness via matrix, we can extract if had worked value decomposition (SVD) exact non-universal since it explicitly depends element-wise distribution P. reliable, statistically optimal but not perfect recovery, universal data-driven algorithm, possible above this using extracted eigen-decompositions. Averaging left right provably improves estimation accuracy threshold. Our results encompass very sparse regime d order 1 SVD fails or produces unreliable recovery. define another variant analysis procedure bypasses randomization step has smaller constant factor computational cost larger polynomial number observed entries. Both thresholds allow go beyond barrier due well-known theoretical limit $$d \asymp \log for found in literature.

برای دانلود باید عضویت طلایی داشته باشید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Preprocessing sparse semidefinite programs via matrix completion

Considering that preprocessing is an important phase in linear programming, it should be systematically more incorporated in semidefinite programming solvers. The conversion method proposed by the authors (SIAM Journal on Optimization, vol. 11, pp. 647–674, 2000, and Mathematical Programming, Series B, vol. 95, pp. 303–327, 2003) is a preprocessing of sparse semidefinite programs based on matri...

متن کامل

Graph Matrix Completion in Presence of Outliers

Matrix completion problem has gathered a lot of attention in recent years. In the matrix completion problem, the goal is to recover a low-rank matrix from a subset of its entries. The graph matrix completion was introduced based on the fact that the relation between rows (or columns) of a matrix can be modeled as a graph structure. The graph matrix completion problem is formulated by adding the...

متن کامل

Crowdclustering with Sparse Pairwise Labels: A Matrix Completion Approac

Crowdsourcing utilizes human ability by distributing tasks to a large number of workers. It is especially suitable for solving data clustering problems because it provides a way to obtain a similarity measure between objects based on manual annotations, which capture the human perception of similarity among objects. This is in contrast to most clustering algorithms that face the challenge of fi...

متن کامل

Crowdclustering with Sparse Pairwise Labels: A Matrix Completion Approach

Crowdsourcing utilizes human ability by distributing tasks to a large number of workers. It is especially suitable for solving data clustering problems because it provides a way to obtain a similarity measure between objects based on manual annotations, which capture the human perception of similarity among objects. This is in contrast to most clustering algorithms that face the challenge of fi...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

ژورنال

عنوان ژورنال: Foundations of Computational Mathematics

سال: 2022

ISSN: ['1615-3383', '1615-3375']

DOI: https://doi.org/10.1007/s10208-022-09568-6