TCN-MAML: A TCN-based Model with Model-agnostic Meta-learning for Human-to-Human Interaction Recognition

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/93034

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/93034

Title:	TCN-MAML: A TCN-based Model with Model-agnostic Meta-learning for Human-to-Human Interaction Recognition
Authors:	林家佑;Lin, Chia-Yu
Contributors:	資訊工程學系
Keywords:	無線訊號;行為辨識;時間卷積網路;小樣本學習;Wi-Fi;Channel State Information;Human Activity Recognition;Temporal Convolution Network;Few-shot learning
Date:	2023-07-11
Issue Date:	2024-09-19 16:39:06 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來，人類活動識別（HAR）的問題引起了相當大的關注，尤其是在使用基於Wi-Fi的應用中，如醫療保健（例如呼吸和心率監測）、安全（例如身份驗證及入侵偵測）、老人照護等方面。對於基於Wi-Fi的HAR而言，這種日益增長的興趣來自於其在不同領域中提供的潛在益處和多功能性。然而，Wi-Fi信號容易受到干擾，在不同時間、不同環境和不同受試者之間波動不定。此外，在基於Wi-Fi的領域中，很少有大量且豐富的資料集，這使得「訓練一個通用模型並透過Wi-Fi訊號識別新的人類活動」變得困難。其中一個主要的解決方案是元學習，它使模型能夠在僅經過少數步驟的情況下適應新的任務。為了應對上述挑戰，我們提出了一種新的方法，將時間卷積網絡（TCN）與模型不可知元學習（MAML）相結合。值得注意的是，我們提出的方法展示了出色的計算效率，同時實現了更高的準確性，即使處理樣本數有限的資料集也是如此。通過對一個公開可訪問的資料集進行嚴格的實驗，我們的方法取得了顯著的結果，展示了令人印象深刻的98.35％的準確性，同時有效地適應新的受試者，並凸顯了它在應對不同場景中的多功能性和韌性。 ;The issue of human activity recognition (HAR) has garnered significant attention in recent years, especially in the utilization of Wi-Fi-based application, such as healthcare (e.g., monitoring breath and heart rate), security, elderly care, and more. This growing interest stems from the potential benefits and versatility offered by Wi-Fi-based HAR in diverse domains. However, the Wi-Fi signals fluctuate through time, environments, and the subjects. Also, there is nearly an extensive dataset in Wi-Fi-based field which leads training a general model to recognize new human activity through Wi-Fi signals difficultly. One main solution is meta-learning which enables the model to adapt to new tasks with only few steps. In order to address the aforementioned challenges, we present a novel approach that combines the Temporal Convolution Network (TCN) with Model-Agnostic Meta-Learning (MAML). Notably, our proposed approach demonstrates remarkable computational efficiency while achieving improved accuracy, even when dealing with datasets that have a limited number of samples. By conducting rigorous experiments on a publicly accessible dataset, we have obtained remarkable results with our approach, showcasing an impressive accuracy of 98.35% while adapting effectively to new subjects and highlighting its versatility and robustness in handling varying scenarios.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	14	View/Open

社群 sharing

Loading...