Research indexConference paper
Angle-Based Multi-Class Kernelized Support Tensor Machine
Zhang
24 Apr 2026Open access
NobleIDark:/48914/w1/20261105/88B6C91FDOI10.1145/3829441.3829466
Angle-Based Multi-Class Kernelized Support Tensor Machine
Zhang · 2026
NobleIDark:/48914/w1/20261105/88B6C91FDOI10.1145/3829441.3829466
Free to read at the publisher
You can read the abstract here, or the full text at the publisher.
Abstract
The widespread application of high-dimensional tensor data in fields such as medical imaging and computer vision has rendered its multi-class classification task a research focus in supervised tensor learning. How to preserve structural information and mine the discriminative nonlinear relationships of tensor data is critical to enhancing the performance of learning tasks. In this paper, we propose an innovative angle-based multi-class kernelized support tensor machine(AM-KSTM). This model integrates kernelized tensor decomposition with angle constraints. Specifically, by introducing kernelized decomposition techniques to approximate tensor data in the kernel space, it can effectively capture the complex nonlinear relationships inherent in the data. The classification task is accomplished by comparing the magnitudes of the included angles between the target tensor and class vectors. Meanwhile, an efficient alternating optimization algorithm is designed to reduce the computational complexity of high-order tensor decomposition. Experiments on the BraTS2018 and BMTDA brain tumor MRI tensor datasets show that AM-KSTM significantly outperforms five mainstream methods including vectorized SVM, CP decomposition+SVM, 3D-CNN, etc in classification accuracy. Moreover, it exhibits better interpretability than 3D-CNN as well as strong robustness.