Beyond the Signs: Nonparametric Tensor Completion via Sign Series
arXiv:2102.00384
Abstract
We consider the problem of tensor estimation from noisy observations with possibly missing entries. A nonparametric approach to tensor completion is developed based on a new model which we coin as sign representable tensors. The model represents the signal tensor of interest using a series of structured sign tensors. Unlike earlier methods, the sign series representation effectively addresses both low- and high-rank signals, while encompassing many existing tensor models -- including CP models, Tucker models, single index models, several hypergraphon models -- as special cases. We show that the sign tensor series is theoretically characterized, and computationally estimable, via classification tasks with carefully-specified weights. Excess risk bounds, estimation error rates, and sample complexities are established. We demonstrate the outperformance of our approach over previous methods on two datasets, one on human brain connectivity networks and the other on topic data mining.
arXiv comment28 pages, 6 figures
References in corpus (10)
- Square Deal: Lower Bounds and Improved Relaxations for Tensor Recovery
- Generalized Canonical Polyadic Tensor Decomposition
- A Consistent Histogram Estimator for Exchangeable Graph Models
- Learning tensors from partial binary measurements
- Provable Convex Co-clustering of Tensors
- Multiway clustering via tensor block models
- Analyzing Tensor Power Method Dynamics in Overcomplete Regime
- Bayesian Network--Response Regression
- Tensor denoising and completion based on ordinal observations
- On Learning High Dimensional Structured Single Index Models