Mitigating Exposure Bias in Risk-Aware Time Series Forecasting with Soft Tokens

2025-12-10

AI Summary

To address exposure bias in autoregressive models—caused by teacher forcing during training—that undermines stability in closed-loop multi-step forecasting and elevates clinical risk in diabetes and hemodynamic management, this paper proposes the Soft Token Trajectory Prediction (SoTra) framework. SoTra replaces discrete tokens with continuous probability distributions for trajectory propagation and integrates a risk-aware decoding module that jointly optimizes a risk-weighted loss and calibrates predictive uncertainty to minimize expected clinical harm. Evaluated on glycemic prediction, SoTra reduces regional clinical risk by 18%; on blood pressure prediction, it lowers effective clinical risk by approximately 15%. These improvements significantly enhance the reliability and robustness of safety-critical time-series forecasting and closed-loop control systems.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Sequential Decision Making
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

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.

Problem

• Mitigates exposure bias in autoregressive time series forecasting
• Learns calibrated uncertainty-aware trajectories for clinical risk
• Minimizes expected clinical harm in safety-critical predictive control

Innovation

• Soft tokens propagate continuous probability distributions
• Risk-aware decoding minimizes expected clinical harm
• Method reduces exposure bias in autoregressive forecasting