IDEAgent: Agentic Quality-Diversity Search for Research Idea Generation

Researchers introduced IDEAgent, a multi-agent system that reframes research idea generation as a quality-diversity search problem.

It combines repair and refinement steps to improve idea quality, while memory-based mechanisms help keep ideas distinct from one another.

The team also proposed Yield, a metric for counting how many high-quality ideas are mutually diverse.

Across multiple computer science topics, IDEAgent outperformed prior LLM-based ideation methods by a wide margin.