In June 2025, at the International Symposium on Computer Architecture (ISCA)—one of the top-tier academic conferences in computer architecture—and the 4th International Workshop on Quantum-Classical Collaborative Computing (QCCC-25), a collaborative research effort by QureGenAI, Ningbo University of Engineering, and China Pharmaceutical University was presented. The paper, titled "SMILES-Inspired Transfer Learning for Quantum Operator Pools in Generative Quantum Eigensolver,"was delivered as a keynote presentation. The paper sparked extensive discussion and received high recognition during the symposium. Notably, QureGenAI was the only Chinese team awarded a Contributed Talk at the event. Recently, an updated version of this research has been published on arXiv, marking a significant breakthrough in the integration of quantum computing and artificial intelligence.
This study pioneers a "transfer learning framework for quantum operator representation based on the Generative Quantum Eigensolver (GQE)."By drawing inspiration from the representation learning ideas of SMILES (Simplified Molecular Input Line Entry System) in cheminformatics, the research applies transfer learning techniques to optimize quantum operator pools in the Generative Quantum Eigensolver. The core innovations include: 1.Chemically Inspired Operator Representation Learning Leveraging the sequential representation characteristics of SMILES strings, the research team constructed an embedded representation space for quantum operators. A pre-trained model was used to learn the semantic features of the operators, forming a transferable operator representation library. This approach captures functional correlations between quantum operators, significantly improving the efficiency of operator selection. 2.Dynamic Operator Pool Optimization Mechanism Traditional methods like ADAPT-VQE require iterative selection of optimal operators from a fixed operator pool, which incurs high computational costs. This study introduces a dynamic operator pool construction mechanism via transfer learning, utilizing a pre-trained model to rapidly generate customized operator sequences tailored to specific molecular systems. This avoids the brute-force search process inherent in conventional methods. 3.Knowledge Transfer Across Molecular Systems By transferring operator representation knowledge from previously trained molecular systems to new ones, a "warm-start" optimization is achieved. Experimental results demonstrate that this method maintains high computational accuracy, ensures consistency of similar features across molecules, and significantly reduces the number of iterations and computational time required for convergence. It is particularly suitable for quantum chemistry problems such as electronic structure calculations and molecular simulations. This research represents a major advancement at the intersection of quantum computing and biomedicine, offering a new paradigm for quantum algorithm design in the Noisy Intermediate-Scale Quantum (NISQ) era. It opens new pathways for the application of quantum computing in drug design, materials science, and related fields.
Paper Address: https://arxiv.org/abs/2509.19715
Repository Address: https://github.com/QureGenAI-Biotech/TyxonQ/tree/main/resear...
TyxonQ Website: https://github.com/QureGenAI-Biotech/TyxonQ/