arXiv subjects

Genghui Li

Publications and source records attributed to Genghui Li.

3 recordsLinked to original sources

arXiv2026

PACE: Primitive-Aware Code Evolution for Automated Algorithm Design

Large Language Model (LLM)-based automated algorithm design typically evolves algorithms as complete, indivisible programs. While this whole-program perspective simplifies the search space, it fundamentally couples the useful local logic to its host program. Consequently, valuable code snippets vanish when the overall program is discarded, making it highly difficult to assess the contribution of individual algorithmic components.To address this, we propose Primitive-Aware Code Evolution (PACE), which decouples local logic from complete programs by representing it as persistent units called Executable Algorithmic Primitives (EAPs). To enable code-level transfer, PACE maintains a dynamic set of EAPs. Algorithm evolution is driven by primitive-aware operators that structurally guarantee the retention and cross-program transfer of these components. To evaluate them effectively, PACE leverages Thompson sampling based on parent-relative performance improvements, guiding primitive selection from the set without requiring extra evaluation datasets. Experiments on four tasks demonstrate that PACE effectively discovers competitive algorithms while structurally preserving valuable algorithmic components.

cs.SE↗

arXiv2025

Large Language Model-Driven Surrogate-Assisted Evolutionary Algorithm for Expensive Optimization

Surrogate-assisted evolutionary algorithms (SAEAs) are a key tool for addressing costly optimization tasks, with their efficiency being heavily dependent on the selection of surrogate models and infill sampling criteria. However, designing an effective dynamic selection strategy for SAEAs is labor-intensive and requires substantial domain knowledge. To address this challenge, this paper proposes LLM-SAEA, a novel approach that integrates large language models (LLMs) to configure both surrogate models and infill sampling criteria online. Specifically, LLM-SAEA develops a collaboration-of-experts framework, where one LLM serves as a scoring expert (LLM-SE), assigning scores to surrogate models and infill sampling criteria based on their optimization performance, while another LLM acts as a decision expert (LLM-DE), selecting the appropriate configurations by analyzing their scores along with the current optimization state. Experimental results demonstrate that LLM-SAEA outperforms several state-of-the-art algorithms across standard test cases. The source code is publicly available at https://github.com/ForrestXie9/LLM-SAEA.

cs.NE↗

arXiv2022

Cross-Domain Few-Shot Relation Extraction via Representation Learning and Domain Adaptation

Few-shot relation extraction aims to recognize novel relations with few labeled sentences in each relation. Previous metric-based few-shot relation extraction algorithms identify relationships by comparing the prototypes generated by the few labeled sentences embedding with the embeddings of the query sentences using a trained metric function. However, as these domains always have considerable differences from those in the training dataset, the generalization ability of these approaches on unseen relations in many domains is limited. Since the prototype is necessary for obtaining relationships between entities in the latent space, we suggest learning more interpretable and efficient prototypes from prior knowledge and the intrinsic semantics of relations to extract new relations in various domains more effectively. By exploring the relationships between relations using prior information, we effectively improve the prototype representation of relations. By using contrastive learning to make the classification margins between sentence embedding more distinct, the prototype's geometric interpretability is enhanced. Additionally, utilizing a transfer learning approach for the cross-domain problem allows the generation process of the prototype to account for the gap between other domains, making the prototype more robust and enabling the better extraction of associations across multiple domains. The experiment results on the benchmark FewRel dataset demonstrate the advantages of the suggested method over some state-of-the-art approaches.

cs.CL↗