Figure 5.

It shows the Temporal-Aware Attention Mechanism (TAM) within the AGOS framework. AGOS uses dynamic attention to capture important phases in disease progression across time-series medical data. Its temporal attention mechanism assigns varying importance weights to time steps, ensuring precise representation of patient histories. The framework integrates domain-constrained optimization with entropy regularization and clinical priors for better interpretability and alignment with clinical knowledge. Temporal-aware self-attention refines temporal dependencies, enabling robust modeling of long-range interactions, significantly enhancing predictive performance in healthcare applications.