EpiEvolve: Self-Evolving Agents for Streaming Pandemic Forecasting under Regime Shifts

2026-06-03 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing, Data Science & Analytics · Depth: Expert, quick

Summary

EpiEvolve is a self-evolving agent designed for streaming pandemic forecasting, specifically addressing the mismatch between static Large Language Model (LLM) forecasters and the dynamic nature of operational pandemic forecasting, which involves continuous data streams and shifting disease regimes. The system wraps an LLM forecaster, keeping its weights fixed, and adapts by storing forecast outcomes in a hierarchical episodic memory. It reflects on delayed labels, retrieves cases relevant to the current regime, and distills recurring errors into strategic rules. This context allows the forecaster to reuse its own past predictions and outcomes chronologically, preventing future leakage. In weekly COVID-19 hospitalization trend forecasting across five variant regimes, EpiEvolve achieved 0.629 average accuracy, significantly outperforming the static backbone (0.561) and the external CDC ensemble (0.325). It also reduced recovery lag after regime shifts from 5 to 2 weeks, with ablations confirming the contributions of reflection, strategic memory, and regime-aware retrieval.

Key takeaway

For Machine Learning Engineers developing adaptive forecasting models for dynamic environments, EpiEvolve demonstrates that self-evolving agent architectures can significantly enhance performance. You should consider integrating hierarchical episodic memory, reflection on delayed labels, and regime-aware retrieval into your LLM-based systems. This approach reduces recovery lag after regime shifts and improves accuracy, moving beyond static supervised models for streaming data challenges.

Key insights

EpiEvolve improves streaming pandemic forecasting by enabling LLMs to self-evolve through episodic memory, reflection, and strategic rule distillation.

Principles

• Operational forecasting requires adaptive, streaming models.
• Fixed-weight LLMs can adapt via external memory and reflection.
• Regime shifts demand dynamic, context-aware adaptation.

Method

EpiEvolve stores forecast outcomes in hierarchical episodic memory, reflects on delayed labels, retrieves regime-relevant cases, and distills errors into strategic rules for LLM context.

In practice

• Apply episodic memory for LLM adaptation.
• Implement reflection on delayed labels.
• Use regime-aware retrieval for context.

Topics

• Pandemic Forecasting
• Large Language Models
• Self-Evolving Agents
• Episodic Memory
• Regime Shifts
• COVID-19 Hospitalization

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• EpiEvolve: Self-Evolving Agents for Streaming Pandemic Forecasting under Regime Shifts — LLM forecasters can adapt to streaming data and regime shifts via memory-based self-evolution, not parameter updates.
• EVIL: Evolving Interpretable Algorithms for Zero-Shot Inference on Event Sequences and Time Series with LLMs — LLM-guided evolution can discover compact, interpretable algorithms for zero-shot dynamical systems inference, outperforming complex neural networks.
• GenEvolve: Self-Evolving Image Generation Agents via Tool-Orchestrated Visual Experience Distillation — Self-evolving image generation agents learn from best-worst trajectory comparisons via visual experience distillation for improved tool orchestration.
• Self-Evolving Multi-Agent Framework for Efficient Decision Making in Real-Time Strategy Scenarios — SEMA is a multi-agent LLM framework for RTS that uses structural entropy and hybrid memory to achieve fast, consistent decisions.
• The Epi-LLM Framework: probing LLM behavioral priors through epidemiological agent-based models — The Epi-LLM framework uses LLMs in agent-based models to simulate epidemic behavior, showing LLMs can reduce infection peaks.
• COMAP: Co-Evolving World Models and Agent Policies for LLM Agents — COMAP co-evolves world models and agent policies through closed-loop interaction for adaptive LLM agents.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.