Skip to content
MAGICS Lab

ASME Journal of Mechanical Design · 2025

Resource Governance in Networked Systems via Integrated Variational Autoencoders and Reinforcement Learning

Govern a multi-agent network by acting in a learned latent space, not on the raw graph.

Qiliang Chen, Babak Heydari

Variational autoencoder applied to network topology — original graph is encoded into a Gaussian latent space, then decoded into a reconstructed graph.
Variational autoencoder applied to network topology — original graph is encoded into a Gaussian latent space, then decoded into a reconstructed graph.

How do you govern a multi-agent network when the space of possible network structures is astronomically large? This work pairs a Variational Autoencoder with deep reinforcement learning, letting a system manager learn to reshape the network by acting in a compact latent space rather than on the raw graph.

Evaluated in modified OpenAI particle environments, the approach beats baselines and surfaces interpretable strategies for balancing system performance against the resource cost of intervention.