Scalable and Interpretable Brain-Inspired Hyper-Dimensional Computing Intelligence with Hardware-Software Co-Design

During the advancement of modern deep learning algorithms, models become increasingly demanding in computing resources and power-hungry, such that they are considered less hardware-friendly for many real-world deployments. The motivation behind brain-inspired computing, or neuromorphic computing, is that the human brain remains the most sophisticated yet efficient learning module ever. We focus on HyperDimensional Computing (HDC), which aims to realize efficient learning via brain-like high-dimensional vector operations. Prior research works have shown that HDC is a lightweight alternative to deep learning in various applications, such as classification and reinforcement learning. HDC can also serve as a reasoning machine on graph datasets and an efficient information retrieval method for genomic sequencing. In this paper, we revisit hardware-software codesigns of HDC, covering the latest developments in both HDC algorithms and accelerator designs. We also carried out extensive comparisons between HDC works and the state-of-the-art.