What question did this study set out to answer?

The aim is to develop a framework that efficiently extracts meaningful representations from various wearable physiological signals.

March 26, 2026

Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals

Key Result

The NormWear foundation model for wearable physiological signals outperformed baselines across 11 datasets and 18 health applications under zero-shot, partial-shot, and full-shot settings.

Key Points

The aim is to develop a framework that efficiently extracts meaningful representations from various wearable physiological signals.
Proposed NormWear framework for multi-modal wearable sensing data.
Implemented a channel-aware attention mechanism with a shared CLS token.
Pretrained on diverse physiological signals from public datasets.
Evaluated across 11 public datasets covering 18 applications.
Demonstrated exceptional generalizability across multiple datasets.
Outperformed competitive models in zero-shot, partial-shot, and full-shot settings.
Applicable in diverse real-world health applications such as mental health and vital sign estimation.

Structured PICO

Does the NormWear foundation model improve performance across diverse wearable sensing applications compared to baselines?

Population

Physiological signals (PPG, ECG, EEG, GSR, IMU) from various public datasets

Intervention

NormWear foundation model

Comparator

Competitive baselines

Outcome

Model performance and generalizability across 11 public wearable sensing datasets

The NormWear foundation model provides a generalized framework for extracting informative representations from diverse wearable sensing data, outperforming existing baselines across multiple health applications.

Main Result

Absolute Event Rate: 0% vs 0%

Abstract

Time-series foundation models excel at tasks like forecasting across diverse data types by leveraging informative waveform representations. Wearable sensing data, however, pose unique challenges due to their variability in patterns and frequency bands, especially for healthcare-related outcomes. The main obstacle lies in crafting generalizable representations that adapt efficiently across heterogeneous sensing configurations and applications. To address this, we propose NormWear , the first paradigm multi-modal and ubiquitous framework of foundation model designed to extract generalized and informative representations from wearable sensing data. Specifically, we design a channel-aware attention mechanism with a shared special liaison CLS token to detect signal patterns in both intra-sensor and inter-sensors. This helps the model to extract more meaningful information considering both time series themselves and the relationships between input sensors. This helps the model to be widely compatible with various sensors settings. NormWear is pretrained on a diverse set of physiological signals, including PPG, ECG, EEG, GSR, and IMU, from various public datasets. Our model shows exceptional generalizability across 11 public wearable sensing datasets, spanning 18 applications in mental health, body state inference, vital sign estimation, and disease risk evaluation. It consistently outperforms competitive baselines under zero-shot, partial-shot, and full-shot settings, indicating broad applicability in real-world health applications.

Bookmark