Characterization of Training Circuits for Hybrid Quantum-classical Algorithms
Sukin Sim, Harvard University
Performing useful computations with current and near-term quantum computers is becoming increasingly viable due to rapid advances in both algorithms and hardware. One class of algorithms, hybrid quantum-classical (HQC), shows promise for demonstrating such near-term utility. A common ingredient or subroutine that plays a crucial role in the algorithmic performance of many of these HQC frameworks is the structure of a parameterized quantum circuit used to train the quantum state optimally or near-optimally for a particular application. Ultimately, having a deeper understanding of the qualities associated with effective parameterized circuits can lead to significant improvements in the performance of HQC algorithms and advance our use of quantum technology. In this work, we introduce a framework for characterizing and evaluating a set of parameterized circuits by defining several descriptors, including a measure of a circuit's representability such that given a particular problem to solve using a HQC algorithm, we have a systematic way of selecting a circuit that is well-suited for the problem. Following the definitions, we demonstrate their utility by computing those quantities for a select set of parameterized circuits and show variances among circuits, highlighting the importance of quantum circuit design.
Abstract Author(s): Sukin Sim, Peter D. Johnson, Alan Aspuru-Guzik