Oxford researchers have introduced Recurrent Structural Policy Gradient (RSPG), the first history-aware hybrid structural method for partially observable mean field games.

Key Contributions

**10x faster convergence** than previous methods
- **State-of-the-art performance**
- **MFAX framework**: JAX-based open-source implementation

## The Problem

Mean Field Games model large populations of interacting agents. At scale, population dynamics become deterministic, but partially observable settings have been challenging.

## The Solution

RSPG combines:
- Monte Carlo rollouts for common noise
- Exact estimation conditioned on samples
- Recurrent architecture for history awareness

## Applications

- Economics and market modeling
- Multi-agent AI systems
- Population dynamics simulation
- Resource allocation

GitHub

Code available at: github.com/CWibault/mfax