Southwest Quantum Information and Technology

Real-time characterization of time-dependent noise with deep reinforcement learning

Presenting Author: Danial Khosravani, Georgia Institute of Technology
Contributing Author(s): Kenneth R. Brown

NISQ-era quantum processors are afflicted by both time-dependent Markovian and non-Markovian noise. Characterizing the time-dependent noise on a quantum computer can help with designing optimal 1,2-qubit gates, improving quantum error-correction codes and increasing the depth of quantum circuits on near-term quantum devices. Common methods for noise characterization are not designed for a general time-dependent noise and the existing quantum spectrum analyzer methods are not suitable for real-time characterization. Here we formulate the problem as a partially observed Markov decision processes (POMDP) and utilize deep reinforcement learning to construct a qubit which can learn a general time-dependent noise model by real-time adaptation of dynamical decoupling and Ramsey measurements. We show that our model is capable of real-time learning of Lindblad equation. Finally we extend this to two qubits and show that the method can be used to obtain spatially correlated time-dependent noise in real-time. We conduct simulations for various scenarios and provide optimality bounds for the specific case of time-dependent noise with bounded two-point correlation.

(Session : from 5:00pm - 7:00pm)