Mitigating Exposure Bias in Risk-Aware Time Series Forecasting with Soft Tokens
By: Alireza Namazi, Amirreza Dolatpour Fathkouhi, Heman Shakeri
Potential Business Impact:
Helps doctors predict health problems better.
Autoregressive forecasting is central to predictive control in diabetes and hemodynamic management, where different operating zones carry different clinical risks. Standard models trained with teacher forcing suffer from exposure bias, yielding unstable multi-step forecasts for closed-loop use. We introduce Soft-Token Trajectory Forecasting (SoTra), which propagates continuous probability distributions (``soft tokens'') to mitigate exposure bias and learn calibrated, uncertainty-aware trajectories. A risk-aware decoding module then minimizes expected clinical harm. In glucose forecasting, SoTra reduces average zone-based risk by 18\%; in blood-pressure forecasting, it lowers effective clinical risk by approximately 15\%. These improvements support its use in safety-critical predictive control.
Similar Papers
Accelerating Time Series Foundation Models with Speculative Decoding
Machine Learning (CS)
Speeds up predictions for websites and apps.
From Values to Tokens: An LLM-Driven Framework for Context-aware Time Series Forecasting via Symbolic Discretization
Machine Learning (CS)
Predicts future events by turning numbers into words.
Adaptive Conformal Prediction Intervals Over Trajectory Ensembles
Machine Learning (CS)
Makes predictions more accurate about future events.