Researchers have developed two novel frameworks, MOSAIC and ACEvo, that leverage large language models (LLMs) for automated heuristic design in combinatorial optimization problems. These systems adversarially co-evolve problem instances and heuristic solvers, creating dynamic evaluation environments that push the boundaries of algorithm performance. Unlike static training methods, MOSAIC and ACEvo generate challenging instances that expose solver weaknesses, leading to more robust and specialized heuristics. The resulting portfolios of heuristics consistently outperform existing LLM-based automated heuristic design methods and achieve broader coverage of problem instance features. AI
IMPACT These frameworks could accelerate the development of highly specialized AI agents for complex optimization tasks.
RANK_REASON The cluster contains two research papers detailing novel frameworks for automated heuristic design using LLMs.
- alphaXiv
- CatalyzeX Code Finder for Papers
- Combinatorial optimization problems
- DagsHub
- Gotit.pub
- Hugging Face
- Large language models
- MOSAIC
- Quality Diversity
- ScienceCast
- CatalyzeX
- COPs
- LLMs
- Ruibo Duan
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
