Low-Rank tensor regression: Scalability and applications

Abstract

With the development of sensor and satellite technologies, massive amount of multiway data emerges in many applications. Low-rank tensor regression, as a powerful technique for analyzing tensor data, attracted significant interest from the machine learning community. In this paper, we discuss a series of fast algorithms for solving low-rank tensor regression in different learning scenarios, including (a) a greedy algorithm for batch learning; (b) Accelerated Low-rank Tensor Online Learning (ALTO) algorithm for online learning; (c) subsampled tensor projected gradient for memory efficient learning.