Signal or Noise? Understanding Generative Models for Real-World Sensor Time Series (SensorGen)

Summary

SensorGen (UCLA HAIL Lab, Yuzhe Yang group) is a fully open, large-scale benchmark study of generative modeling for real-world sensor time series. It unifies previously fragmented time-series generation work into one testbed: 14 generation settings, 7 real-world datasets, 12 signal modalities, and 5 generative model families, evaluated across 4 core capabilities: semantic-to-signal generation, interpolation/forecasting, cross-channel translation, and signal editing. It is not a foundation model; it is a map of which generative approaches work for which sensor-signal regimes, starting from human physiology.