AutoREM: Memory-Augmented LLMs for Robust Optimization Reformulation
A novel system known as Automated Reformulation with Experience Memory (AutoREM) leverages memory-augmented large language models to streamline the conversion of robust optimization (RO) problems into manageable deterministic forms. Although RO is a well-founded method for making decisions amid uncertainty, its widespread use is limited by the necessity for manual reformulation, which demands accurate multi-step reasoning and mathematically sound transformations. To facilitate comprehensive assessment, the team also introduced AutoRO-Bench, a benchmark that includes an automated data generation pipeline for essential RO reformulation tasks along with a curated dataset for RO applications. AutoREM operates without the need for tuning, autonomously creating structured experience memory to enhance reformulation precision. This research is detailed in arXiv preprint 2605.11813.
Key facts
- AutoREM is a memory-augmented LLM framework for robust optimization reformulation.
- Robust optimization reformulation requires precise multi-step reasoning.
- AutoRO-Bench is a new benchmark for evaluating LLM-based RO reformulation.
- AutoRO-Bench includes an automated data generation pipeline.
- AutoREM is tuning-free and builds experience memory autonomously.
- The research is published as arXiv:2605.11813.
- RO reformulation is challenging due to need for mathematical consistency.
- LLMs have shown promise for automating optimization formulation.
Entities
Institutions
- arXiv