噪音(视频)
声学
物理
计算机科学
人工智能
图像(数学)
作者
Kees Koenders,Leo Schnitzspan,Fabian Kammerbauer,Sinan Shu,G. Jakob,Mathias Kläui,Johan H. Mentink,Nasir Ahmad,Marcel A. J. van Gerven
标识
DOI:10.1103/physrevapplied.23.054035
摘要
Physical learning machines promise to overcome the von Neumann bottleneck by implementing highly energy-efficient in-situ adaptation. This adaptation requires parameter updates that are local in space and time, while being robust to the inherent noise in physical substrates. This study embraces physical noise generated by stochastic magnetic tunnel junctions as a mechanism for learning via a combination with a recent local noise-based learning rule. The authors demonstrate that learning based on physical noise is a viable strategy, scalable to larger systems and a variety of physical substrates.
科研通智能强力驱动
Strongly Powered by AbleSci AI