Video recognition by physical reservoir computing in magnetic materials

TL;DR

This paper explores the use of nonlinear spin dynamics in magnetic materials as a physical reservoir computing system for video recognition, demonstrating its potential for developing magnetic-based visual sensors.

Contribution

It introduces a practical implementation of magnetic physical reservoirs for video recognition using a novel spatiotemporal parallelization scheme.

Findings

Achieved accurate classification of images in a video recognition task.

Demonstrated the feasibility of magnetic physical reservoirs for visual sensing.

Paved the way for magnetic-based visual sensor development.

Abstract

Nonlinear spin dynamics in magnetic materials offers a promising avenue for implementing physical reservoir computing, one of the most accomplished brain-inspired frameworks for information processing. In this study, we investigate the practical utility of magnetic physical reservoirs by assessing their performance in a video recognition task. Leveraging a recently developed spatiotemporal parallelization scheme, our reservoir achieves accurate classifications of previously provided images. Our findings pave the way for the development of visual sensors based on the magnetic physical reservoir computing.