# WavesFM: Hierarchical Representation Learning for Longitudinal Wearable Sensor Waveforms Source: [WavesFM: Hierarchical Representation Learning for Longitudinal Wearable Sensor Waveforms]() Collection index: [llms.txt]() Data updated: 2026-05-13T05:00:12.693Z Authors: Peng Cao, Zhijian Yang, Tennison Liu, Jonathan Wang, Jiang Wu, Magdalena Proszewska, Arvind Pillai, Mingwu Gao, Amir Farjadian, Lawrence Cai, Emily Blanchard, Daniel McDuff, Pramod Rudrapatna, Matthew Thompson, Anupam Pathak, Mark Malhotra, Shwetak Patel, Dina Katabi, Paolo Di Achille, Ming-Zher Poh Published: 2026-05-09T21:22:02.000Z ## Abstract Wearable sensors enable the continuous acquisition of high-resolution physiological waveforms, such as photoplethysmography and accelerometry, under free-living conditions. However, inferring health-related phenotypes from these signals presents significant challenges due to high sampling frequencies, multimodal dependencies, and extreme sequence lengths (e.g., weeks of recordings), compounded by a scarcity of ground-truth labels. To address these challenges, existing self-supervised learning (SSL) methodologies typically follow two paradigms: (1) learning rich morphological representations from short waveform segments while collapsing longitudinal dynamics through simple aggregation, or (2) modeling behavioral patterns from coarse, hand-crafted features (e.g. heart rate, step counts) spanning longer horizons but foregoing subtle, predictive signatures in raw waveforms. To bridge this gap, we propose WavesFM, a foundation model utilizing a two-stage SSL framework for longitudinal physiological data. Specifically, we decompose the learning problem into two stages: first, a segment-level encoder is pretrained to extract local embeddings from short waveforms; subsequently, a temporal encoder is trained to model the sequence of these embeddings across a multi-day horizon. This hierarchical approach overcomes the computational complexity of high-resolution, long-sequence data, allowing the overall model to capture both local signal semantics and the complex circadian and inter-day variations governing physiological dynamics. Pretrained on over 6.8M hours (N=324k individuals) of recordings for the first stage and 5.3M hours (N=10k) for the second stage, WavesFM demonstrates superior performance across 58 diverse tasks spanning demographics, lifestyle, health conditions, and medications. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Health Conditions Classification on Longitudinal Wearable Sensor Waveforms](): Health Conditions Classification; Longitudinal Wearable Sensor Waveforms; AUROC: 0.89. - [Treatment Classification on Longitudinal Wearable Sensor Waveforms](): Treatment Classification; Longitudinal Wearable Sensor Waveforms; AUROC: 78.4. - [Lifestyle Classification on Longitudinal Wearable Sensor Waveforms](): Lifestyle Classification; Longitudinal Wearable Sensor Waveforms; AUROC: 78.1. - [Demographics Classification on Longitudinal Wearable Sensor Waveforms](): Demographics Classification; Longitudinal Wearable Sensor Waveforms; AUROC: 0.992. - [Active Smoker Prediction on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Active Smoker Prediction; Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 78.4. - [Condition Diagnosis (Diabetes) on Longitudinal Wearable Sensor Waveforms WavesFM (Downstream Tasks)](): Condition Diagnosis (Diabetes); Longitudinal Wearable Sensor Waveforms WavesFM (Downstream Tasks); AUROC: 83.6. - [Condition Diagnosis (Stroke) on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Condition Diagnosis (Stroke); Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 73.6. - [Lifestyle Factor Prediction (Frequent Sugar) on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Lifestyle Factor Prediction (Frequent Sugar); Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 69.7. - [Treatment Identification (Diuretics) on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Treatment Identification (Diuretics); Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 0.788. - [Treatment Identification (Nitrates) on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Treatment Identification (Nitrates); Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 0.864. - [Condition Diagnosis (Atrial Fibrillation) on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Condition Diagnosis (Atrial Fibrillation); Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 0.823. - [Condition Diagnosis (Cancer) on Longitudinal Wearable Sensor Waveforms WavesFM](): Condition Diagnosis (Cancer); Longitudinal Wearable Sensor Waveforms WavesFM; AUROC: 0.718. - [Condition Diagnosis (COPD) on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Condition Diagnosis (COPD); Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 73.3. - [Obesity Prediction on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Obesity Prediction; Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 0.928. - [Older Age Prediction on Longitudinal Wearable Sensor Waveforms WavesFM (Downstream Tasks)](): Older Age Prediction; Longitudinal Wearable Sensor Waveforms WavesFM (Downstream Tasks); AUROC: 0.959. - [Sex Male Prediction on Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM](): Sex Male Prediction; Longitudinal Wearable Sensor Waveforms Downstream Tasks WavesFM; AUROC: 0.996. ## Original sources and implementation [Original paper]() [PDF]()