Neurally-Inspired Hyperdimensional Computing for Robust and Real-time Learning*

In the realm of Applied Physics Science, the contemporary challenges surrounding the efficient execution of learning tasks in computing systems are of paramount concern. Our contribution to this field presents an innovative computing paradigm, inspired by the intricacies of the human brain, designed to address these challenges. This novel computing system not only facilitates various learning and cognitive tasks but also offers a remarkable enhancement in computational efficiency and resilience when compared to existing platforms.

Our platform leverages the concept of HyperDimensional (HD) computing, an unconventional approach to computation that draws inspiration from the fundamental principles underlying brain functionality. HD computing's core objective is to achieve high-efficiency and noise-tolerant computation. It does so by emulating crucial aspects of the human memory model through vector operations. These operations are both computationally feasible and founded on a sound mathematical basis, providing an accurate description of human cognition.

One noteworthy advantage of HD computing lies in its ability to learn from data in one or only a few iterations, often referred to as one-shot or few-shot learning. This means that the system can acquire knowledge from a minimal set of examples and process the training data in a single pass, as opposed to many iterative processes. Furthermore, HD computing exhibits remarkable fault tolerance, as it operates over random hypervectors that are independently and identically distributed. This research presents a compelling avenue for addressing the intricate challenges encountered in modern computing systems, aligning with the objectives and interests of the Applied Physics Science community.