Risk sensitive evolutionary approach improves heuristic algorithm reliability
CVaR-RMD: CVaR Selection Pressure for Robust Metaheuristic Discovery
Artificial Intelligence
Summary
Using average scores to pick computer programs that solve problems can hide occasional big failures. The authors created a new method, CVaR-RMD, which focuses on the worst-case outcomes rather than just averages. This approach uses a large language model to tweak algorithms and selects them by looking at the risk of bad performance across different problem types. Their method reduces how often these algorithms fail while still doing well on average. This improvement holds true even on bigger and different test problems.
What this means in practice
- •For optimization software developers: Create more reliable heuristic algorithms that maintain good performance while reducing rare severe failures across diverse problem types.
- •For supply chain planners: Use risk-aware algorithm selection to improve the robustness of routing and scheduling solutions over varying scenarios and problem scales.
Authors
Ruodai Cui, Li Niu
Abstract
Large language models have recently been used to discover heuristics through iterative program search, offering a promising approach to automated algorithm design. However, candidates selected by average performance may degrade across instance families, random seeds, and problem scales because mean-based objectives can obscure rare but severe failures. We propose CVaR-RMD, a risk-sensitive evolutionary framework that uses an LLM to mutate executable heuristic programs and ranks candidates by empirical conditional value-at-risk over heterogeneous evaluation conditions. Across three combinatorial optimization domains, CVaR-based selection improves upper-tail performance and reduces failure rates while maintaining competitive mean performance. These gains persist on larger unseen instances and independently curated within-domain benchmarks