Agents23h ago
GraphDC Uses Multi-Agent Divide-and-Conquer For Graphs
GraphDC decomposes complex input graphs into smaller subgraphs for specialized agents to process locally. A master agent then integrates these local outputs with inter-subgraph data to solve the overall problem. This hierarchical approach targets the reasoning failures LLMs face with large-scale topologies. It offers a scalable template for graph-based algorithmic tasks.