Quantum Algorithms Meet AI for Quantum State Preparation

VQE suffers from several limitations: the circuit depth can be excessive, and optimization landscapes are plagued by barren plateaus. ADAPT-VQE improves upon this by iteratively constructing problem-specific ansätze from a pool of operators, guaranteeing compact circuits and incorporating hardware constraints directly into the cost function. Yet, ADAPT-VQE is computationally expensive and does not scale well beyond ~15 qubits due to the exponential growth in gradient evaluations.

In this talk, we introduce a new AI-driven approach to quantum state preparation that addresses the limitations of current algorithms for near-term quantum devices. We leverage CUDA-Q with GPU-accelerated classical simulator to generate synthetic training data, enabling efficient learning of quantum circuit patterns and heuristics. This hybrid framework facilitates scalable training of AI models that can generalize across problem instances, paving the way for practical quantum algorithm synthesis.