AUTHOR=Zhang Wenli , Wang Yaoyuan , Ji Xinglong , Wu Yujie , Zhao Rong 

TITLE=ROA: A Rapid Learning Scheme for In-Situ Memristor Networks

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=Volume 4 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2021.692065

DOI=10.3389/frai.2021.692065

ISSN=2624-8212

ABSTRACT=Memristor shows great promise in neuromorphic computing owning to high-density integration, fast computing and low-energy consumption. However, the non-ideal synaptic weight update in memristor devices, including nonlinearity, asymmetry and device variation, still poses challenges to the in-situ learning of memristor, thereby limiting its broad applications. Although existing offline learning schemes can avoid this problem by transferring the weight optimization process into cloud, it is difficult to adapt to unseen tasks and uncertain environments. To alleviate the non-ideal update problem and achieve fast adaptation, we propose a bi-level learning scheme with high accuracy, which effectively combines offline pre-training and online rapid one-step adaption with a regularization constrain and a dynamic learning rate. Furthermore, the model has been implemented on memristor devices to solve few-shot learning tasks and demonstrates superiority over pure offline and online schemes under noisy conditions. Our hybrid on-chip method can solve in-situ learning in imprecise memristor networks, providing potential applications of on-chip neuromorphic learning and edge computing.