Transformers for Time Series Data

Faculty: Zach Ives

with Jiaming Liang, Lei Cao, Sam Madden, and Guoliang Li

Many key health biomarkers are captured in timeseries data:

  • blood pressure, heart rate, EKG, EEG, CRP, …

We need artificial intelligence to handle this at scale:

  • Can we train AI to detect important diagnostic or predictive events in time, across multi-modal and multi-channel timeseries data?
  • Can we automatically cluster related phenomena in real-time?
  • Can we build databases and digital biobanks of timeseries
    symptoms, and instantly find similar patterns?

Our RITA algorithm uses transformers to address these tasks over multi-channel timeseries – and is currently being investigated for epileptic seizure prediction.


In progress: building a data platform that supports integration of embeddings and database data.