IDEAgent: Agentic Quality-Diversity Search for Research Idea Generation

Researchers introduced IDEAgent, a multi-agent framework that jointly optimizes research idea quality and diversity.

It evolves ideas through lineage-based search, combining repair and refinement with diversity-preserving memory and comparisons.

They also proposed Yield, a metric that counts how many diverse ideas clear a quality threshold.

Across 32 topics in 8 computer science domains, IDEAgent outperformed prior baselines by a wide margin.