Retrieval-Augmented Personalization with Foundation Models for Wearable Stress Detection

2026-06-23

AI Summary

This work addresses the challenge of personalizing stress detection in wearable devices due to inter-individual physiological variability by proposing a lightweight, retrieval-augmented approach that requires no fine-tuning. Leveraging a frozen foundation model, the method retrieves historical physiological segments from the target user that exhibit similar patterns and constructs a compact personalized embedding to modulate the representations of a lightweight Transformer. Relying solely on unlabeled multimodal physiological signals—including electrodermal activity (EDA), blood volume pulse (BVP), temperature, and accelerometer data—the approach achieves a 3.92% absolute improvement in accuracy and a 4.76% gain in macro F1 score on the WESAD dataset, closely matching the performance of supervised fine-tuning. The study further demonstrates the feasibility of temporal retrieval and cross-dataset personalization.

Abstract

Personalization in wearable-based stress detection remains challenging due to substantial inter-individual variability in physiological and behavioral responses. While traditional approaches rely on user-specific fine-tuning or costly self-supervised pre-training on large datasets, we propose a lightweight alternative based on retrieval-augmented personalization. Our method leverages frozen, out-of-domain foundation models to retrieve similar patterns from a target user's history and encode them into a compact personalized embedding that modulates representations extracted by a lightweight transformer network. We evaluate our approach on the WESAD stress detection dataset with N=15 users, comprising wrist-worn physiological (EDA, BVP, temperature) and activity (accelerometer) signals, and report gains of +3.92\% in accuracy and +4.76\% in macro F1-score over a non-personalized transformer baseline, approaching supervised fine-tuning performance without requiring any labeled user data. We further show that temporal retrieval, where only prior user samples are available, achieves performance close to full intra-user retrieval, demonstrating robustness to limited user history. Finally, we explore personalization in a cross-dataset retrieval setting, leveraging embeddings from the K-Emocon dataset to personalize representations for stress detection on the WESAD dataset.

Problem

• personalization
• stress detection
• wearable devices
• inter-individual variability
• foundation models

Innovation

• retrieval-augmented personalization
• foundation models
• wearable stress detection
• personalized embedding
• few-shot adaptation