PulseImpute: A Novel Benchmark Task for Pulsative Physiological Signal Imputation

The promise of Mobile health (mHealth) is the ability to use wearable sensors to monitor a person's physiology at high frequencies during daily life to enable temporally-precise health interventions. However, a major challenge is frequent missing data. Despite a rich imputation literature, existing techniques are ineffective for the pulsative signals which comprise many mHealth applications. Moreover, the lack of large-scale datasets with labeled examples of missingness have prevented the ML community from tackling this important problem. We address this gap with PulseImpute, the first mHealth signal imputation challenge which includes realistic missingness models, an extensive set of baselines, and clinically-relevant downstream tasks that describe signal structures. The well-defined signal structure brings an additional emphasis on accurate shape reconstruction, and we demonstrate that existing state-of-the-art methods fail, adversely affecting their downstream clinical applications. We hypothesize that these models are unable to effectively exploit the quasi-periodic signal property for imputation, and thus, as a proof-of-concept, we introduce an augmented self-attention mechanism that attends on quasi-periodic features and achieves strong performance. We hope that PulseImpute will enable the ML community to tackle this significant and challenging task.