Hyper-dimensional computing architectures for AI applications: A comprehensive survey of principles, performance, and future prospects

This paper discusses different Hyper-Dimensional Computing (HDC) architectures and their utilization in solving various artificial intelligence (AI) applications. HDC is based on the principles of high-dimensional vector spaces and presents a distinct way of representing and processing information. The paper first presents an outline of the fundamental principles, mathematical underpinnings, and essential operations of HDC, thereby establishing the necessary basis for subsequent discussions. Afterward, the paper explores prominent conventional and modern HDC architectures, clarifying their operational principles and practical uses. In addition, the paper also discusses the performance evaluation of different HDC architectures, in terms of accuracy, efficiency, and scalability, in solving real-world tasks. The limitations of existing HDC architectures were also examined in this paper, and their plausible remedies are proposed. Lastly, the paper provides an analysis of future prospects, encompassing research trends to fully utilize HDC as an alternative to deep learning to solve different classification and recognition tasks. Offering a bird’s-eye perspective of HDC architectures and their changing place in AI, this paper presents a comprehensive survey for academics, industry professionals, and interested parties working on AI-based applications.