From Mixing to Tearing: Graph Decomposition in Decentralized Optimization via Message Passing
The paper proposes GATE, a decentralized optimization framework that decomposes graph structure into cooperative message-passing subproblems.
Ding and Scutari target smooth strongly convex optimization over undirected networks where each agent can only talk to neighbors. Their method jointly designs agreement constraints, dual-variable blocks, and connected agent clusters, rather than treating communication as a separate mixing layer. GATE assigns one variable per edge and solves tree-block subproblems through endpoint cost-to-go messages; GATE-S trades exactness for cheaper local surrogate updates. The authors prove linear convergence with a rate tied to function regularity, topology, and partition choice, and report numerical tests supporting the theory. Source: ArXiv · AI/CL/LG's note.
Ding and Scutari target smooth strongly convex optimization over undirected networks where each agent can only talk to neighbors. Their method jointly designs agreement constraints, dual-variable blocks, and connected agent clusters, rather than treating communication as a separate mixing layer. GATE assigns one variable per edge and solves tree-block subproblems through endpoint cost-to-go messages; GATE-S trades exactness for cheaper local surrogate updates. The authors prove linear convergence with a rate tied to function regularity, topology, and partition choice, and report numerical tests supporting the theory. Source: ArXiv · AI/CL/LG's note.
score 4