Controlling Graph Dynamics with Reinforcement Learning and Graph Neural Networks
This addresses the challenge of efficient intervention in dynamic graph-based systems like epidemics or social networks, but appears incremental as it builds on existing RL and GNN methods.
The paper tackles the problem of controlling partially-observed dynamic processes on graphs with limited interventions, such as curbing epidemics or maximizing influence, by formulating it as a sequential decision problem and designing a tractable scheme. The approach is successfully applied to epidemic testing prioritization and influence maximization, though no concrete numbers are provided in the abstract.
We consider the problem of controlling a partially-observed dynamic process on a graph by a limited number of interventions. This problem naturally arises in contexts such as scheduling virus tests to curb an epidemic; targeted marketing in order to promote a product; and manually inspecting posts to detect fake news spreading on social networks. We formulate this setup as a sequential decision problem over a temporal graph process. In face of an exponential state space, combinatorial action space and partial observability, we design a novel tractable scheme to control dynamical processes on temporal graphs. We successfully apply our approach to two popular problems that fall into our framework: prioritizing which nodes should be tested in order to curb the spread of an epidemic, and influence maximization on a graph.