EVENT DOSSIER
The researchers proposed a new framework called QuantumEvo, aimed at using large language models as heuristic generators to optimize the quantum circuit synthesis process based on binary decision diagrams (BDD). This framework involves searching for variable ordering strategies initialized by various heuristic families, directly manipulating variable order using standard BDD operations, and filtering candidate solutions using downstream quantum cost (QCC) metrics. The HGA-QE heuristic method modifies the screening steps in genetic algorithms to more effectively align with quantum costs. In the benchmark dataset, HGA-QE achieved a win-rate of 70.9% and was strictly superior to the single-function best baseline in 13.5% of functions, demonstrating broad competitive performance and relative advantages on test sets built from different data sources.
研究人员提出名为 QuantumEvo 的框架，利用大语言模型（LLM）作为启发式生成器，针对基于二进制决策图（BDD）的量子电路综合进行量子成本（QCC）感知的变量排序优化。该框架通过搜索由多种启发式家族初始化的排序策略，利用标准 BDD 操作直接操纵变量顺序并由下游 QCC 筛选候选者。其中发现的 HGA-QE 启发式修改了遗传算法中的筛分步骤以更好地对齐 QCC。在基准测试集中，HGA-QE 获得 70.9% 的平局或胜率，并在 13.5% 的功能上严格优于单函数最佳基线，展现出广泛的竞争性能及在不同数据源构建的测试套件上的相对优势。