Practical quantum optimization is limited by qubit counts, circuit depth, hardware noise, and the high cost of variational parameter training. We present DQAOA-GPT, a framework that integrates artificial intelligence, quantum computing, and high-performance computing (AI–QC–HPC) to overcome these barriers. At its core, QAOA-GPT, a generative pretrained transformer, directly generates problem-adapted QAOA circuits and eliminates variational optimization entirely. Distributed QAOA (DQAOA) decomposes large problems into subproblems that are executed in parallel across quantum–classical resources, which enables optimization well beyond the capacity of a single quantum device. We further embed DQAOA-GPT in an active learning loop. A factorization machine surrogate is converted into a binary optimization model and solved by DQAOA-GPT. Selected solutions are then evaluated and fed back to refine the surrogate. We demonstrate this framework on materials design problems with increasing dimensionality and higher-order interactions. Together, training-free circuit generation and distributed execution substantially reduce the computational cost of large-scale optimization. This provides a scalable path toward integrated AI–quantum–HPC workflows for scientific discovery.
Chicago, IL 60616
United States