OR-Agent: Bridging Evolutionary Search and Structured Research for Automated Algorithm Discovery
TLDR
OR-Agent is a multi-agent framework combining evolutionary search and structured research trees for automated algorithm discovery.
Reasoning
The paper introduces a novel integration of evolutionary and systematic ideation with optimization-inspired reflection, validated on combinatorial optimization and cooperative driving simulations. However, it focuses narrowly on algorithm discovery and does not address literature review or paper writing.
Read-first score
Read-first score 73.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 71.
Field roles
Rank sensitivity
Stability: volatile; rank range: 30.
Keyword Scores
Deep Analysis
Innovations
- Structured tree-based workflow with branching hypothesis generation and systematic backtracking for controlled research trajectory management
- Evolutionary-systematic ideation mechanism that unifies evolutionary selection of starting points, comprehensive research plan generation, and coordinated exploration within a research tree
- Hierarchical optimization-inspired reflection system using short-term reflections as verbal gradients, long-term reflections as verbal momentum, and memory compression as semantic weight decay
Methodology
OR-Agent is a configurable multi-agent framework that organizes research as a tree-based workflow with branching and backtracking. It combines an evolutionary-systematic ideation mechanism with a hierarchical reflection system inspired by optimization concepts. Experiments are conducted on classical combinatorial optimization benchmarks and simulation-based cooperative driving scenarios, comparing against strong evolutionary baselines.
Key Results
OR-Agent outperforms strong evolutionary baselines on both combinatorial optimization benchmarks and cooperative driving scenarios, demonstrating a general and inspectable framework for AI-assisted scientific discovery.