Learning robust and high-precision quantum controls

Abstract

Robust and high-precision quantum control is extremely important but challenging for the functionization of scalable quantum computation. In this paper, we show that this hard problem can be translated to a supervised machine learning task by treating the time-ordered quantum evolution as a layer-ordered neural network (NN). The seeking of robust quantum controls is then equivalent to training a highly generalizable NN, to which numerous tuning skills matured in machine learning can be transferred. This opens up a door through which a family of robust control algorithms can be developed. We exemplify such potential by introducing the commonly used trick of batch-based optimization, and the resulting stochastic b-GRAPE algorithm is numerically shown to be able to remarkably enhance the control robustness while maintaining high fidelity.

Optimized three-level quantum transfers based on frequency-modulated optical excitations

“…The increase of control parameters can then be exploited for fine-tuned shaping of the effective Hamiltonian. Possible objectives of this procedure may be the minimization of the oscillations around the target adiabatic path 45 , or the robustness against selected types of external noise 46,47 . In this direction, the method may be combined with successful techniques from optimal control for carrying-out the optimization task, such as stochastic gradient (learning) algorithms 47 .…”

“…Possible objectives of this procedure may be the minimization of the oscillations around the target adiabatic path 45 , or the robustness against selected types of external noise 46,47 . In this direction, the method may be combined with successful techniques from optimal control for carrying-out the optimization task, such as stochastic gradient (learning) algorithms 47 . All the mentioned features of fmod-STIRAP make it particularly fascinating from the experimental point of view, showing its potential for generalizations to speed up adiabatic passages in more-level configurations, and hence for successful applications in many branches of quantum science.…”

Learning Control of Quantum Systems

“…The performance of SLC approach can be further improved by exploring the richness and diversity of samples. Inspired by deep learning, a batchbased gradient algorithm (b-GRAPE) has been presented for efficiently seeking robust quantum controls, and numerical results showed that b-GRAPE can achieve improved performance over the SLC method for remarkably enhancing the control robustness while maintaining high fidelity (Wu et al 2019). In other applications where we need to enhance the robustness in closed-loop learning control, we may either use the Hessian matrix information (Xing et al 2014) or integrate the idea of SLC into the learning algorithm in searching for robust control fields.…”

An improved differential evolution algorithm for learning high-fidelity quantum controls

“…This difficulty is further exacerbated when taking into account of control uncertainties. One can employ machine learning methods [15][16][17][18][19][20][21] to deal with this problem. Another promising approach to tackle this problem is to apply gradient-free algorithms, for example, Nelder-Mead algorithm (NM) [22,23] and differential evolution algorithm (DE) [5,6,[24][25][26].…”