Signal or Noise? Understanding Generative Models for Real-World Sensor Time Series
For researchers working with real-world sensor time series, this provides a systematic benchmark and design guidelines for generative modeling, though it is an incremental contribution as it applies existing methods to a new domain.
The paper introduces SensorGen, a large-scale study of sensor-signal generation across 14 settings, 4 domains, 7 datasets, and 12 modalities. It finds that flow-matching models perform best overall, signal properties like demographic covariates and time-frequency modeling improve generation, and synthetic data enhances downstream performance.
Generative models have changed how machine learning represents complex data distributions, especially in language and vision, yet many real-world systems are observed instead as continuous, high-dimensional, and noisy sensor time series. Existing generative modeling of sensor data, however, remains fragmented across modalities, datasets, and task formulations, limiting a systematic understanding of when, how, and why generative models succeed or fail in real-world settings. To address this gap, we introduce SensorGen, a large-scale study of sensor-signal generation spanning 14 settings across 4 domains, 7 datasets, and 12 signal modalities. Leveraging SensorGen, we systematically evaluate generative models from five major families and uncover three key findings: (1) flow-matching models provide strong overall performance across most settings; (2) signal properties matter, with demographic covariates improving longitudinal generation and time-frequency modeling improving high-frequency signal generation; and (3) generated signals have practical utility beyond visual realism, with scaling improving generation quality and synthetic data improving downstream performance. Together, SensorGen establishes a broader understanding of design choices, evaluation protocols, and failure modes in real-world sensor data generation.