Year: 2024
Type: conference-paper
Abstract: Motivated by the increasing size of the state-of-the-art machine-learning models and their ever-increasing training costs, hyperdimensional computing approaches have gained massive traction. While those do not beat other models’ accuracy, they provide advantageous training and inference times. Arguably, it is the training costs that hinder the ML research. In this work, we propose a new computationally cheap approach to improve the accuracy of the HDC models. We verify our proposed method on multiple different datasets. Despite being unable to beat more computationally-heavy approaches, we believe that the experimental results show a promising trend in its applicability.
Authors: Matej Cibula, Radek Marik
Institution: Czech Technical University in Prague
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4411747387