Machine Learning for Reliable Quantum Computing

Summary: Reliable quantum computing requires co-design across the full stack, including algorithm structure, compilation, and execution, rather than isolated optimization of any single layer. This talk presents a unified framework in which these layers are jointly optimized under realistic hardware constraints. Structure-aware compilation exploits algorithmic interaction graphs to reduce circuit depth; joint optimization across mapping, term ordering, and algorithm parameters yields significantly improved success probability; and execution-layer analysis reveals that error mitigation effectiveness depends critically on circuit symmetry and native gate choice. While demonstrated on superconducting quantum processors, the underlying co-design principles are potentially transferable across architectures. Building on this foundation, the talk highlights machine learning as the key enabler for closing the feedback loop: data-efficient, parameterized noise models learned directly from workload execution data achieve substantially improved predictive fidelity over vendor models without the overhead of dedicated characterization. These insights situate the co-design perspective within the emerging framework of quantum deep learning. In this broader setting, the talk also presents an operational definition, a unifying taxonomy, and a benchmarking protocol grounded in explicit resource contracts.