SOTER improves wearable health signal predictions with spectral and time modeling
SOTER: A Generative Time-Series Foundation Model for Wearable Human Physiological Signals
Machine LearningArtificial Intelligence
Summary
Wearable devices collect complex health signals that are noisy and irregularly timed, making analysis difficult. The authors developed SOTER, a new computer model that understands relationships across different signals and their rhythms over time. By training on billions of data points, SOTER can predict missing information and classify health states more accurately than existing methods. This model works well even when data is noisy or incomplete. The work shows how specialized models can better handle the unique challenges of wearable health data.
What this means in practice
- •For wearable device developers: Improve health signal forecasting and gap-filling in wearable device data streams under real-world noisy and irregular sampling conditions.$Commercial implications: Enables development of advanced wearable health devices that offer better continuous monitoring by accurately predicting and imputing physiological signals.
- •For healthcare analytics teams: Use a pretrained model to classify patient physiological states from wearable sensor data without retraining for new datasets.
Authors
Fangke Chen, Sirry Chen, Wei Chen, Zhongyu Wei
Abstract
Time-series foundation models have demonstrated strong cross-domain transfer, yet their common architectural assumptions remain poorly aligned with wearable physiological signals, which are multichannel, irregularly sampled, noisy, and governed by coupled continuous-time dynamics spanning distinct spectral scales. We present SOTER, a generative foundation model for wearable physiological time series that unifies cross-channel coupling, spectrum-guided expert specialization, and continuous-time latent evolution within a single pre-training framework. SOTER combines a spatial feature-aware backbone that models inter-signal dependencies, a power spectral density (PSD)-guided mixture-of-experts layer that routes representations to experts associated with fixed spectral bands through an inspectable, non-learned rule, and a neural controlled differential equation decoder that supports prediction and imputation at arbitrary timestamps. We pre-train SOTER on 226 billion time points from five public physiological datasets and evaluate the same pre-trained model across out-of-distribution zero-shot forecasting, frozen-encoder linear-probe classification, and continuous-time imputation on wearable benchmarks. SOTER achieves the best RMSE on 4 of 6 datasets and the best MAE on 5 of 6 in zero-shot forecasting, the highest average Macro-AUROC in classification, and the lowest imputation error on all six datasets at 75% missingness. It further remains robust to additive acquisition noise, matching or surpassing baselines evaluated on clean inputs even under the strongest corruption. These results indicate that domain-specialized foundation models for wearable physiology benefit from jointly modeling channel structure, spectral scale, and continuous-time dynamics.