Abstract:Attribute to its powerful representation ability, block term decomposition (BTD) has recently attracted many views of multi-dimensional data processing, e.g., hyperspectral image unmixing and blind source separation. However, the popular alternating least squares algorithm for rank-( L , M , N ) BTD (BTD-ALS) suffers expensive time and space costs from Kronecker products and solving low-rank approximation subproblems, hindering the deployment of BTD for real applications, especially for large-scale data. In this paper, we propose a fast sketching-based Kronecker product-free algorithm for rank-( L , M , N ) BTD (termed as KPF-BTD), which is suitable for real-world multi-dimensional data. Specifically, we first decompose the original optimization problem into several rank-( L , M , N ) approximation subproblems, and then we design the bilateral sketching to obtain the approximate solutions of these subproblems instead of the exact solutions, which allows us to avoid Kronecker products and rapidly solve rank-( L , M , N ) approximation subproblems. As compared with BTD-ALS, the time and space complexities and of KPF-BTD are significantly cheaper than and of BTD-ALS, where . Moreover, we establish the theoretical error bound for KPF-BTD. Extensive synthetic and real experiments show KPF-BTD achieves substantial speedup and memory saving while maintaining accuracy (e.g., for a synthetic tensor, the running time 0.2 seconds per iteration of KPF-BTD is significantly faster than 96.2 seconds per iteration of BTD-ALS while their accuracies are comparable).

An Improved Frequent Directions Algorithm for Low-Rank Approximation via Block Krylov Iteration

Accelerating Exact Nearest Neighbor Search in High Dimensional Euclidean Space Via Block Vectors

Robust Frequent Directions with Application in Online Learning

Effective Streaming Low-Tubal-Rank Tensor Approximation via Frequent Directions

Frequent Direction Algorithms for Approximate Matrix Multiplication with Applications in CCA.

Near Optimal Frequent Directions for Sketching Dense and Sparse Matrices

A Randomized Block Krylov Method for Tensor Train Approximation

Efficient Randomized Algorithms for the Fixed-Precision Low-Rank Matrix Approximation

Towards Efficient and Accurate Approximation: Tensor Decomposition Based on Randomized Block Krylov Iteration

Online Learning Via Regularized Frequent Directions.

Learning-Based Low-Rank Approximations

More on low rank approximation of a matrix

On the Unreasonable Effectiveness of Single Vector Krylov Methods for Low-Rank Approximation

Efficient Orthogonal Decomposition with Automatic Basis Extraction for Low-Rank Matrix Approximation

Randomized algorithms for low-rank matrix approximation: Design, analysis, and applications

A Fast Algorithm for Rank-(L, M, N) Block Term Decomposition of Multi-Dimensional Data

Enhancing RBF-FD Efficiency for Highly Non-Uniform Node Distributions via Adaptivity

A Rank-One Update Algorithm for Fast Solving Kernel Foley–Sammon Optimal Discriminant Vectors

Low-Rank Matrix Iteration Using Polynomial-Filtered Subspace Extraction.

Robust Blockwise Random Pivoting: Fast and Accurate Adaptive Interpolative Decomposition

Randomized Douglas-Rachford methods for linear systems: Improved accuracy and efficiency