Aggregation of Evolutionary Game Dynamics on Large-Scale Weighted Networks

What the math gives to ML

The paper provides an exact backward-equivalence aggregation for weighted networked evolutionary games: agents in the same class have identical future behavior when their weighted interactions with every equivalence class and their local update rules agree. The transferable asset is an equitable-partition and quotient construction that replaces an N-agent dynamical system by a K-class system while preserving trajectories on the class-synchronized invariant subspace. In neural networks, this can become an exact quotient message-passing architecture or a dynamic graph coarsening module, with a falsifiable zero-error condition when node features are class-constant and a measurable residual when the condition is only approximate.