Reproduce “Skill Learning Using Process Mining for Large Language Model Plan Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Andrei Cosmin Redis, Mohammadreza Fani Sani, Bahram Zarrin, Andrea Burattin
A framework integrates process mining techniques to enable text-based Large Language Models to learn and retrieve structured skills, improving plan generation with explicit control flow. This approach demonstrated the feasibility of deriving accurate process models from LLM-generated plans, and a hybrid retrieval method is projected to outperform purely semantic baselines given sufficiently accurate initial LLM plan generation.
Key insights
- Process mining techniques are effective in discovering structured control flow models from LLM-generated action sequences, enabling skill learning for text-based planners.
- The full benefit of structural conformance checking for skill retrieval is realized when the LLM's initial 'thought' generation is sufficiently accurate (e.g., alignment fitness >= 0.7).
- Integrating process models into LLM planning provides enhanced interpretability of generated plans and identifies opportunities for parallel execution, offering benefits beyond just retrieval accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.