DyACE: Dynamic Algorithm Co-evolution for Online Automated Heuristic Design with Large Language Model
arXiv:2603.13344v1 Announce Type: new Abstract: The prevailing paradigm in Automated Heuristic Design (AHD) typically relies on the assumption that a single, fixed algorithm can effectively navigate the shifting dynamics of a combinatorial search. This static approach often proves inadequate for Perturbative Heuristics, where the optimal algorithm for escaping local optima depends heavily on the specific search phase. To address this limitation, we reformulate heuristic design as a Non-stationary Bi-level Control problem and introduce DyACE (Dynamic Algorithm Co-evolution). Distinct from standard open-loop solvers, DyACE use a Receding Horizon Control architecture to continuously co-evolve the heuristic logic alongside the solution population. A core element of this framework is the Look-Ahead Rollout Search, which queries the landscape geometry to extract Search Trajectory Features. This sensory feedback allows the Large Language Model (LLM) to function as a grounded meta-controller,
arXiv:2603.13344v1 Announce Type: new Abstract: The prevailing paradigm in Automated Heuristic Design (AHD) typically relies on the assumption that a single, fixed algorithm can effectively navigate the shifting dynamics of a combinatorial search. This static approach often proves inadequate for Perturbative Heuristics, where the optimal algorithm for escaping local optima depends heavily on the specific search phase. To address this limitation, we reformulate heuristic design as a Non-stationary Bi-level Control problem and introduce DyACE (Dynamic Algorithm Co-evolution). Distinct from standard open-loop solvers, DyACE use a Receding Horizon Control architecture to continuously co-evolve the heuristic logic alongside the solution population. A core element of this framework is the Look-Ahead Rollout Search, which queries the landscape geometry to extract Search Trajectory Features. This sensory feedback allows the Large Language Model (LLM) to function as a grounded meta-controller, prescribing phase-specific interventions tailored to the real-time search status. We validate DyACE on three representative combinatorial optimization benchmarks. The results demonstrate that our method significantly outperforms state-of-the-art static baselines, exhibiting superior scalability in high-dimensional search spaces. Furthermore, ablation studies confirm that dynamic adaptation fails without grounded perception, often performing worse than static algorithms. This indicates that DyACE's effectiveness stems from the causal alignment between the synthesized logic and the verified gradients of the optimization landscape.
Executive Summary
The DyACE algorithm, introduced in the article, presents a novel method for dynamic automated heuristic design in combinatorial search problems. By reformulating heuristic design as a non-stationary bi-level control problem and utilizing a receding horizon control architecture, DyACE continuously co-evolves the heuristic logic alongside the solution population. This approach leverages a large language model as a grounded meta-controller, prescribing phase-specific interventions based on real-time search status. The results demonstrate superior scalability in high-dimensional search spaces, outperforming state-of-the-art static baselines. However, ablation studies suggest that dynamic adaptation fails without grounded perception, highlighting the importance of causal alignment between synthesized logic and optimization landscape gradients.
Key Points
- ▸ DyACE reformulates heuristic design as a non-stationary bi-level control problem
- ▸ Utilizes receding horizon control architecture for continuous co-evolution
- ▸ Leverages large language model as grounded meta-controller
- ▸ Demonstrates superior scalability in high-dimensional search spaces
- ▸ Dynamic adaptation fails without grounded perception
Merits
Strength in Scalability
DyACE's ability to adapt to changing search landscapes and outperform static baselines in high-dimensional spaces is a significant merit.
Innovative Use of LLM
The integration of a large language model as a grounded meta-controller is a novel and effective approach to heuristic design.
Demerits
Dependence on Grounded Perception
Dynamic adaptation fails without grounded perception, which may limit the applicability of DyACE in certain scenarios.
Complexity of Implementation
The receding horizon control architecture and large language model integration may add complexity to the implementation of DyACE.
Expert Commentary
The DyACE algorithm presents a significant advancement in automated heuristic design, leveraging the strengths of large language models and co-evolutionary principles. While the results are promising, it is essential to consider the limitations and potential complexities of implementation. Further research should focus on refining the algorithm's adaptability and scalability, as well as exploring its applications in various fields. The article's findings have far-reaching implications for optimization problems and may inform policy decisions in areas like resource allocation and public planning.
Recommendations
- ✓ Future research should prioritize the development of more robust and adaptable optimization heuristics, incorporating advances in large language models and co-evolutionary principles.
- ✓ The DyACE algorithm should be further tested and validated on a broader range of optimization problems to assess its generalizability and scalability.