Noise-based Local Learning using Stochastic Magnetic Tunnel Junctions

2024-12-17

arXiv.org

AI Summary

To address the challenge of physical noise disrupting learning in brain-inspired hardware, this work proposes a noise-driven local learning mechanism. It leverages the intrinsic thermal/tunneling noise of spintronic stochastic magnetic tunnel junctions (s-MTJs) as a natural learning signal source and integrates it with spike-timing-dependent plasticity (STDP)-inspired update rules, enabling distributed training of multilayer spiking neural networks without global error backpropagation. Contrary to conventional noise-suppression paradigms, this approach actively harnesses noise as a learning driver. Experimental validation on small-scale physical neuromorphic systems demonstrates classification accuracy approaching that of standard backpropagation, while achieving over two orders-of-magnitude reduction in energy consumption. This work establishes a novel learning principle and hardware implementation pathway for robust, ultra-low-power physical neural morphic computing.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Bio-inspired Learning
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Social networks and social learning
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Brain-inspired learning in physical hardware has enormous potential to learn fast at minimal energy expenditure. One of the characteristics of biological learning systems is their ability to learn in the presence of various noise sources. Inspired by this observation, we introduce a novel noise-based learning approach for physical systems implementing multi-layer neural networks. Simulation results show that our approach allows for effective learning whose performance approaches that of the conventional effective yet energy-costly backpropagation algorithm. Using a spintronics hardware implementation, we demonstrate experimentally that learning can be achieved in a small network composed of physical stochastic magnetic tunnel junctions. These results provide a path towards efficient learning in general physical systems which embraces rather than mitigates the noise inherent in physical devices.

Problem

• Develop noise-based learning for physical neural networks
• Achieve efficient learning using stochastic magnetic tunnel junctions
• Embrace inherent device noise instead of mitigating it

Innovation

• Noise-based learning in physical neural networks
• Stochastic magnetic tunnel junctions implementation
• Energy-efficient alternative to backpropagation algorithm