Hyperdimensional Computing Regression: Foundations, Algorithms, and Comparative Review

This article offers a targeted survey and comparative analysis of regression techniques based on hyperdimensional computing (HDC), and investigates how they compare with traditional regression and machine-learning approaches. Hyperdimensional regression encompasses a family of methods that first transform data into high-dimensional vectors and subsequently learn a predictive model within that transformed space. A key observation is that many recent approaches attain competitive performance by stepping away from the original associative algebra of HDC and instead applying iterative linear regression, often of the ridge-regression type, after the encoding step. To disentangle pure predictive performance from hardware-related influences, representative methods are assessed in a software-only environment and compared with standard linear, kernel-based, tree-based, and neural network regressors on benchmark datasets. The review underscores the need for more systematic encoder design, strategies for selecting the hyper-dimensionality, robustness assessments, and a renewed investigation of regression schemes that exploit binding and bundling more fully while still supporting efficient hardware realization.