Efficient Quantum Circuit Compression using Reinforcement Learning

No Thumbnail Available
Date
2021-05
Authors
Berthusen, Noah
Major Professor
Advisor
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Authors
Research Projects
Organizational Units
Organizational Unit
Physics and Astronomy
Physics and astronomy are basic natural sciences which attempt to describe and provide an understanding of both our world and our universe. Physics serves as the underpinning of many different disciplines including the other natural sciences and technological areas.
Journal Issue
Is Version Of
Versions
Series
Abstract
Computations by the current generation of noisy intermediate scale quantum (NISQ) computers are often plagued by errors such as decoherence and cross talk. Such errors severely limit the depth of NISQ quantum circuits, yet many quantum algorithms that show promise of a quantum speedup require deep circuits and prolonged coherence times. In this work, we propose leveraging Reinforcement Learning (RL) to intelligently build quantum circuits that can recreate given target states, given no information about the circuit used to construct them. The RL agent learns about the hidden system by receiving rewards based on local observables, calculated using the target state, and the fidelity of the final state. By constraining the depth of the circuits built by the agent, we hypothesize that this approach allows us to compress the depth of quantum circuits necessary to create the target state. One important application of our method is dynamic quantum simulation, where the target state is a time-evolved state using a given Hamiltonian and a Trotterized quantum circuit. Our method promises quantum simulations out to longer final times than are currently feasible on NISQ devices.
Comments
Description
Keywords
Citation
DOI
Source
Subject Categories
Copyright