Robust Control of Quantum-dot Spin Qubits from Machine Learning

Project: Research

View graph of relations


The quantum technology is expected to outperform present-day computational technologies in many important classes of problems. Semiconductor spin qubits are being actively studied as one of the leading candidates to host a quantum computer, and accurate control of spin qubits is key to the realization of quantum computing. Among the various techniques to improve the control accuracy, the dynamically corrected gates--which make noises cancelling themselves--are effective means to reduce gate errors. Composite pulse sequences that correct errors dynamically have been developed for both the singlet-triplet qubit and the exchange-only qubit.On the other hand, more and more laborious tasks are conducted by trained machines thanks to the development of artificial intelligence. Machine learning is a powerful tool that allows data analysis or optimization that is beyond the ability of human being or any enumerative methods previously imagined. In this project, we are going to use two methods from machine learning to the quantum control of spin qubits: supervised learning and reinforcement learning. We will develop techniques to measure the noise spectra of quantum devices using supervised learning, in which a neural network is trained with simulated randomized benchmarking results. We will also apply reinforcement learning to two important classes of problems in manipulating quantum states: quantum state preparation and transfer. Last, with the assistance of reinforcement learning, we shall develop quantum control sequences that are optimized as compared to previously found ones, both in terms of their length and complexity but also regarding to specific functions that we desire. In particular, we aim to develop composite pulse sequences that cancel noises with certain spectra which is beyond the low-frequency noises previously studied.As an outcome of this study, we believe the combination of machine learning and quantum control shall offer us new techniques and understanding in controlling spin qubit systems, which may potentially shed light on the endeavor toward scalable fault-tolerant quantum computation.


Project number9042695
Grant typeGRF
Effective start/end date1/01/198/02/23