Hyperdimensional computing in biomedical sciences: a brief review

Abstract

Hyperdimensional computing (HDC, also known as vector-symbolic architectures-VSA) is an emerging computational paradigm that relies on dealing with vectors in a high-dimensional space to represent and combine every kind of information. It finds applications in a wide array of fields including bioinformatics, natural language processing, machine learning, artificial intelligence, and many other scientific disciplines. Here we introduced the basic foundations of the HDC, focusing on its application to biomedical sciences, with a particular emphasis to bioinformatics, cheminformatics, and medical informatics, providing a critical and comprehensive review of the current HDC landscape, highlighting pros and cons of applying this computational paradigm in these specific scientific domains. In this study, we first selected around forty scientific articles on hyperdimensional computing applied to biomedical data existing in the literature, and then analyzed key aspects of their studies, such as vector construction, data encoding, programming language employed, and other features. We also counted how many of these scientific articles are open access, how many have public software code available, how many groups of authors, journals, and conferences are most present among them. Finally, we discussed the advantages and limitations of the HDC approach, outlining potential future directions and open challenges for the adoption of HDC in biomedical sciences. To the best of our knowledge, our review is the first open brief survey on this topic among the biomedical sciences, and therefore we believe it can be of interest and useful for the readership.

Keywords: Bioinformatics; Biomedical sciences; Cheminformatics; Hyperdimensional computing; Medical informatics; Review; Vector-symbolic architectures.

Figure 1. Number of research articles considered in this survey on hyperdimensional computing applied to biomedical sciences published over time.

Note that 2025 refers only to January and the first half of February.

Figure 2. Distribution of publications over 15 scientific problems discussed in the research articles considered in this survey (A), and the distribution of research articles across the three areas of biomedical sciences, i.e., medical informatics, bioinformatics, and cheminformatics, with a strong prevalence of articles falling into the first category (B).

Figure 3. Distribution of research articles based on the type of random vectors (i.e., binary, bipolar, and unspecified–A), data encoding technique (i.e., record-based, n-gram, record-based and n-gram, and unspecified–B), and data modeling method (MAP model and unspecified or unclear–C) used for building the HDC architecture over the set of manuscripts considered in this study.

Figure 4. Distribution of software implementations over the different types of hardware architectures, programming languages, and software availability.

Distribution of software implementations described in the research articles considered in this article over the different types of hardware architectures (i.e., CPU, FPGA, GPU, ARMC, ASIC, BCI, AIMC, CMOS, TPU, PCM, and unspecified–A) for which they have been specifically designed (note that the same implementation could be available for more than a single type of hardware) together with the distribution of the same number of research articles over the programming languages used to develop the HDC models (B) and the number of articles for which authors made their software available (C).

Figure 5. Bar graph with the distribution of the 62 research articles considered in this study over the publication venues.

(A) Conference proceedings, journals, and preprints; (B) publishers; (C) open access.