Zero-shot forecasting of epidemics.
Advances in Neural Information Processing Systems (NeurIPS) - BERT2S., 2025
Accurate epidemic forecasting is essential for timely interventions, yet traditional models often rely on limited task-specific data. This work evaluates pre-trained Large Language Models (LLMs) for zero-shot time series forecasting of epidemic dynamics across multiple horizons. The study benchmarks LLM-based approaches against classical and deep learning models while analyzing their architectural suitability.
<img src="/images/LLM_TSF.png" alt=LLM TSF" style="width: 90%; max-width: 800px;">