ADAPT-GQE combines transformer-based generative AI with quantum computing to generate molecular ground-state preparation circuits more efficiently than traditional optimization methods, targeting a key bottleneck in quantum chemistry workflows.
The framework trains models on ADAPT-VQE output, then improves them with reinforcement learning on NVIDIA CUDA-Q simulations so they can propose novel circuits rather than just copy the teacher algorithm.
Quantinuum validated the best AI-generated circuits on its Helios hardware, including what it called one of the largest AI-generated quantum chemistry circuits yet run on a quantum computer, using imipramine as the test molecule.
The partners said the proof of principle could support more automated, repeatable drug-discovery workflows, while also offering a path to broader molecular simulation in materials, energy, agriculture and electronics.
Longer term, the team aims to build quantum foundation models trained on increasingly complex quantum data, eventually designing circuits for molecules too large for classical simulation.