CEoH
Contextual Evolution of Heuristics
EoH with a problem description added to every prompt, so smaller LLMs can design heuristics for niche problems.
The loop
- Proposes
Tree-search scoring heuristics for the unit-load pre-marshalling problem, as thought plus code
- Judged by
Running the heuristic and measuring the average relative gap of its move count to a lower bound
- Steers
Fitness drives EoH selection; the added problem description is kept in every prompt
Headline result
For unit-load pre-marshalling, Qwen2.5-Coder:32b with CEoH reached a fitness of 12.5% (best overall), against 14.3% for the best EoH run (GPT-4o 2024-08-06). The "Optimal A*" baseline scored about 8.1%. CEoH outperformed EoH for Gemma2:27b, Qwen2.5-Coder:32b and DeepSeekV3:685b.
What this does not prove
One niche problem. On unseen instances fitness rose to 30.8 (Qwen CEoH) and 33.9 (GPT-4o EoH), which the authors read as overfitting. For GPT-4o EoH was better than CEoH, and none solved the hardest case (80% fill).
How it generalizes the harness
An evolutionary loop over a population produces a heuristic that is plugged into a search algorithm and used without the model.
Classification
| What it proposes | Heuristic |
|---|---|
| Generative model | Pretrained LLMs, frozen (Gemma2:27b, Qwen2.5-Coder:32b, DeepSeekV3:685b, GPT-4o) |
| Algorithmic authority | Instance evaluation of heuristic inside a tree search |
| Feedback | Score |
| What gets updated | Population of candidates |
| When it runs | Offline, at design time |
| Validity | Empirical (measured) |
| Evidence | Experiments reported |
| Lineage | Derivative |