MetaWearS: A Shortcut in Wearable Systems Lifecycle with Only a Few Shots

TL;DR

MetaWearS tackles data scarcity and energy constraints in wearable health monitoring by fusing meta-learning with prototypical updates. The method pretrains on a base dataset using episodic meta-learning to form class prototypes, then updates prototypes with a few new shots from new subjects, enabling accurate deployment on wearables without retraining the full model. Empirical results across epileptic seizure (EEG) and AF (ECG) detection show strong AUC performance with limited target data, and substantial energy savings (up to hundreds of times) for prototype updates on X-HEEP hardware. The work demonstrates practical, low-data, and energy-efficient adaptation suitable for continuous health monitoring in real-world wearable deployments.

Abstract

Wearable systems provide continuous health monitoring and can lead to early detection of potential health issues. However, the lifecycle of wearable systems faces several challenges. First, effective model training for new wearable devices requires substantial labeled data from various subjects collected directly by the wearable. Second, subsequent model updates require further extensive labeled data for retraining. Finally, frequent model updating on the wearable device can decrease the battery life in long-term data monitoring. Addressing these challenges, in this paper, we propose MetaWearS, a meta-learning method to reduce the amount of initial data collection required. Moreover, our approach incorporates a prototypical updating mechanism, simplifying the update process by modifying the class prototype rather than retraining the entire model. We explore the performance of MetaWearS in two case studies, namely, the detection of epileptic seizures and the detection of atrial fibrillation. We show that by fine-tuning with just a few samples, we achieve 70% and 82% AUC for the detection of epileptic seizures and the detection of atrial fibrillation, respectively. Compared to a conventional approach, our proposed method performs better with up to 45% AUC. Furthermore, updating the model with only 16 minutes of additional labeled data increases the AUC by up to 5.3%. Finally, MetaWearS reduces the energy consumption for model updates by 456x and 418x for epileptic seizure and AF detection, respectively.

Figures (13)

• Figure 1: (a) Challenges in a wearable system lifecycle: Initial data collection, further data collection for updates, and inefficient update protocol. (b) The proposed method addresses all the mentioned challenges.
• Figure 2: Meta-testing and Meta-training in few-shot learning, highlighting the creation of episodes with three distinct classes and two samples randomly selected from a base dataset. The meta-testing process involves entirely new classes to form a support set and query set, mirroring the tasks in the meta-training phase.
• Figure 3: Hardware structure in X-HEEP. In the MetaWear system, the updated prototypes are transmitted to the CPU through the BLE module.
• Figure 4: Overview of model architectures for signal classification in wearable health monitoring :(a) The EEG signal processing pathway employs a deep learning-based transformer architecture for Epilepsy detection from EEG signals, (b) The AF detection model utilizes a MobileNetV2 neural network to analyze ECG recordings for real-time classification and (c) after computing the feature vector $f_{\varphi(x)}$ for each query sample, these vectors are embedded into a high-dimensional space. In this visual representation, data points are classified by computing their Euclidean distances to class prototype vectors.
• Figure 5: (a) Pretraining and (b) fine-tuning the model on the meta-train step in MetaWearS. This figure also shows the split of data into support set and query set in each episode.