Generative AI Model for Quantum State Preparation

Building upon recent advances in AI-guided quantum circuit construction, this work focuses on training generative models to synthesize efficient quantum state preparation circuits for near-term quantum devices. Combining quantum circuits generated using NVIDIA's hybrid quantum-classical framework, CUDA-Q, with molecular descriptors, we generate reference data to train a transformer-based generative AI model for quantum state preparation across various configurations of a drug molecule. In this talk, I will discuss the progressive refinement of our generative AI model toward larger and more complex quantum systems. We then compare its performance to more conventional state preparation methods in terms of energy accuracy and circuit depth. Finally, we focus on the model's scalability and execution of the generated circuits on Quantinuum's newest quantum computer, Helios. This work establishes a foundation for scalable, data-driven quantum state preparation to explore detailed reaction kinetics and atomistic models of chemical compounds at large