Abstract:This paper presents a simple yet efficient method for statistical inference of tensor linear forms using incomplete and noisy observations. Under the Tucker low-rank tensor model and the missing-at-random assumption, we utilize an appropriate initial estimate along with a debiasing technique followed by a one-step power iteration to construct an asymptotically normal test statistic. This method is suitable for various statistical inference tasks, including constructing confidence intervals, inference under heteroskedastic and sub-exponential noise, and simultaneous testing. We demonstrate that the estimator achieves the Cramér-Rao lower bound on Riemannian manifolds, indicating its optimality in uncertainty quantification. We comprehensively examine the statistical-to-computational gaps and investigate the impact of initialization on the minimal conditions regarding sample size and signal-to-noise ratio required for accurate inference. Our findings show that with independent initialization, statistically optimal sample sizes and signal-to-noise ratios are sufficient for accurate inference. Conversely, if only dependent initialization is available, computationally optimal sample sizes and signal-to-noise ratio conditions still guarantee asymptotic normality without the need for data-splitting. We present the phase transition between computational and statistical limits. Numerical simulation results align with the theoretical findings.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: **In the tensor completion problem, how to make statistical inferences on the tensor linear form from incomplete and noisy observational data and achieve optimal uncertainty quantification.** Specifically, the paper focuses on the following aspects: 1. **Development of statistical inference methods**: - A simple and efficient method is proposed for making statistical inferences on the tensor linear form using incomplete and noisy observational data. - This method is based on the Tucker low - rank tensor model and the random missing assumption. By using appropriate initial estimates and de - biasing techniques, combined with one - step power iteration, an asymptotically normal test statistic is constructed. 2. **Optimal uncertainty quantification**: - The paper shows that the proposed estimator reaches the Cramér - Rao lower bound on the Riemannian manifold, indicating its optimality in uncertainty quantification. 3. **Study of the gap between statistics and computation**: - The gap between statistics and computation is comprehensively examined, and the influence of initialization on the minimum sample size and signal - to - noise ratio conditions required for accurate inferences is studied. - It is found that when the initialization is independent, the statistically optimal sample size and signal - to - noise ratio are sufficient to achieve accurate inferences; when the initialization is dependent, the computationally optimal sample size and signal - to - noise ratio conditions can still ensure asymptotic normality without data splitting. 4. **Extension to multi - task inferences**: - The method is extended to various inference tasks, including the construction of confidence intervals, simultaneous inferences of multiple linear forms, and inferences under heteroscedastic and sub - exponential noise. Through these works, the paper aims to fill the gaps in existing tensor completion methods in terms of statistical inferences and uncertainty quantification, and provide a reliable statistical inference framework to meet the challenges in practical applications.

Statistical Inference in Tensor Completion: Optimal Uncertainty Quantification and Statistical-to-Computational Gaps

Statistical Inference for Low-Rank Tensors: Heteroskedasticity, Subgaussianity, and Applications

Inference for Low-rank Tensors -- No Need to Debias

Tensor Graphical Model: Non-Convex Optimization and Statistical Inference

Rank-Adaptive Tensor Completion Based on Tucker Decomposition

Tensor Denoising Using Low-Rank Tensor Train Decomposition

Low-Rank and Sparse Enhanced Tucker Decomposition for Tensor Completion

Tensor Completion via Integer Optimization

Low-rank Tensor Estimation via Riemannian Gauss-Newton: Statistical Optimality and Second-Order Convergence

Tensor Methods for Generating Compact Uncertainty Quantification and Deep Learning Models

Low Multilinear Rank Approximation of Tensors and Application in Missing Traffic Data

Tensor cumulants for statistical inference on invariant distributions

Low Tucker rank tensor completion using a symmetric block coordinate descent method

An Iterative Reweighted Method for Tucker Decomposition of Incomplete Multiway Tensors

Tensor-on-Tensor Regression: Riemannian Optimization, Over-parameterization, Statistical-computational Gap, and Their Interplay

Tensor Train Factorization and Completion under Noisy Data with Prior Analysis and Rank Estimation

Tucker Tensor Analysis of Matérn Functions in Spatial Statistics

Randomized Algorithms for the Approximations of Tucker and the Tensor Train Decompositions.

Bayesian Sparse Tucker Models for Dimension Reduction and Tensor Completion

Statistical Inference For Noisy Matrix Completion Incorporating Auxiliary Information

Nonconvex Robust High-Order Tensor Completion Using Randomized Low-Rank Approximation