Review of few-shot learning application in CSI human sensing

• DOI: 10.1007/s10462-024-10812-4
• Publication Date: 2024-07-05T05:01:45Z
• Authors: Zhengjie Wang; Jianhang Li; Wenchao Wang; Zhaolei Dong; Qingwei Zhang; Yinjing Guo
• Keywords: Transfer of learning; Retraining

Abstract Wi-Fi sensing has garnered increasing interest for its significant advantages, primarily leveraging signal fluctuations induced by human activities and advanced neural network algorithms. However, application faces challenges due to limited generalizability, necessitating frequent data recollection retraining adaptation new environments. To address these limitations, some researchers introduced few-shot learning into applications because it offers a promising solution with ability achieve remarkable performance in novel scenarios using minimal training samples. Despite potential, comprehensive review of within this domain remains absent. This study endeavors fill gap exploring prominent that incorporate learning, aiming delineate their key features. We categorize approaches three distinct methodologies: transfer metric meta-learning, based on strategies. Through classification, we examine representative systems from an perspective elucidate the principles implementation. These are evaluated terms methodology, modality, recognition accuracy. Finally, paper highlights future directions Channel State Information (CSI) sensing, providing valuable resource field learning.