GraphDC: Multi-Agent Framework for Scalable Graph Algorithm Reasoning
GraphDC is a multi-agent framework that employs a divide-and-conquer strategy to enhance the performance of large language models (LLMs) on tasks involving graph algorithms. Given the intricate nature of graphs, which necessitate thorough multi-step reasoning, particularly when scaled, GraphDC breaks down an input graph into smaller subgraphs. Each subgraph is handled by a dedicated agent for localized reasoning, while a master agent synthesizes the outputs along with inter-subgraph data. This structured approach lessens the reasoning load, mitigates computational limitations, and boosts overall robustness. Comprehensive experiments demonstrate that GraphDC consistently surpasses current methods.
Key facts
- GraphDC is a divide-and-conquer multi-agent framework for scalable graph algorithm reasoning.
- It decomposes an input graph into smaller subgraphs.
- Each subgraph is assigned to a specialized agent for local reasoning.
- A master agent integrates local outputs with inter-subgraph information.
- The framework reduces reasoning burden on individual agents.
- It alleviates computational bottlenecks.
- It improves robustness on large graph instances.
- Extensive experiments show consistent outperformance.
Entities
—