Large Language Models (LLMs) have demonstrated great potential in automating the generation of Verilog hardware description language code for hardware design. This automation is critical to reducing human effort in the complex and error-prone process of hardware design. However, existing approaches predominantly rely on human intervention and fine-tuning using curated datasets, limiting their scalability in automated design workflows. Although recent iterative search techniques have emerged, they often fail to explore diverse design solutions and may underperform simpler approaches such as repeated prompting. To address these limitations, we introduce EvoVerilog, a novel framework that combines the reasoning capabilities of LLMs with evolutionary algorithms to automatically generate and refine Verilog code. EvoVerilog utilizes a multiobjective, population-based search strategy to explore a wide range of design possibilities without requiring human intervention. Extensive experiments demonstrate that EvoVerilog achieves state-of-the-art performance, with pass@10 scores of 89.1 and 80.2 on the VerilogEval-Machine and VerilogEval-Human benchmarks, respectively. Furthermore, the framework showcases its ability to explore diverse designs by simultaneously generating a variety of functional Verilog code while optimizing resource utilization.

Automated Machine Learning (AutoML) has become an essential tool for democratizing machine learning (ML) by automating key aspects of model selection, hyperparameter tuning, and feature engineering [1, 2]. However, the efficiency of AutoML frameworks remains a significant challenge, as the search for optimal configurations is often computationally expensive [3-5]. Traditional search strategies, such as Random Search (RS) and Bayesian Optimization (BO), indiscriminately explore large search spaces, resulting in high resource consumption [3, 6, 7]. To address this challenge, we propose a metalearning approach that dynamically designs search spaces for an AutoML solution, reducing computational costs while maintaining competitive predictive performance. The proposed method leverages historical metaknowledge to identify and prioritize promising regions of the search space, enabling more efficient optimization. By predicting the performance of preprocessor-classifier combinations, a meta-model, induced using metalearning, can provide a warm-start advantage, accelerating the AutoML search process. This study evaluates the effectiveness of the proposed approach through an extensive set of experiments, analyzing both computational efficiency and predictive performance. According to the experimental results, the dynamically generated search spaces significantly reduce runtime, while maintaining high-quality solutions. In particular, the RS-mtl-95 configuration achieved an 89% reduction in runtime without compromising predictive performance.

Neural Information Processing Systems http://nips.cc/

We propose a novel, efficient search method through a general, structured kernel space.

P ( x) could be nonconvex.

Imitation learning aims to extract high-performance policies from logged demonstrations of expert behavior.

Bayesian optimization involves "inner optimization" over a new-data acquisition