Extracting and evaluating performance of NISQ Optimization experiments: beyond angle-parameter setting

The NISQ (Noisy Intermediate-Scale Quantum) era was ushered in approximately 5 years ago, marking a period of experimentation with noisy quantum processors. It has been characterized by a global effort involving hundreds of small-scale experiments across various experimental platforms. These experiments have validated numerous noise models and have tested the practical aspects of quantum heuristics, including QAOA and its simple variants. Owing to advancements in hardware quality and control software, there have been several recent demonstrations of experimental runs on gate-model noisy quantum processors, showcasing the use of over 20 qubits in regimes where simulations become challenging. In this talk, we will discuss insights gained from the DARPA ONISQ program, where NASA, USRA, and Rigetti Computing employed an array of techniques to combat noise while aiming to solve non-trivial binary optimization problems. We will delve into the impacts of the discovered techniques, which encompass ansatz approximations, swap-network synthesis, over-parametrization, categorical parameters like ordering and symmetry transformations, and iterative decompositions. We will also explore how these can be amalgamated into a cohesive algorithm-tuning strategy that can be executed on the fly, achieving high approximation ratios within a few thousand runs for problems with 50+ variables.