Quantum Error Correction using Variational Neural Annealing
Examensarbete för masterexamen
Complex adaptive systems (MPCAS), MSc
Ever since the theoretical possibility of quantum computers was discovered an effort to create the practical possibility has been ongoing. The challenge in creating a quantum computer lies mostly in the extremely sensitive and error prone quantum bits (qubits) making up the basis of the quantum computation. One of the most promising approaches to solve the issues of the sensitive qubits is quantum error correction, which aims to correct for errors rather than eliminating them altogether. One way to do error correction is to create a logical qubit consisting of multiple physical qubits, where errors on a physical qubit can be corrected to preserve the logical qubit from errors. The most common way to achieve this is by implementing a two-dimensional surface code, where a grid of qubits represent a single logical qubit. In order to decode this surface code and correct for errors we need to measure error syndromes on the surface code and decide which qubit error is most likely to cause that syndrome. In this thesis a new and alternative solution to decode with an algorithm called Variational Neural Annealing (VNA) [Hibat-Allah et al, Nature Machine Intelligence, 3, 952 (2021)] is investigated. This an optimization technique that uses a Recurrent Neural Network (RNN). Its viability as a decoder is determined by its accuracy and runtime. Both a two-dimensional and a one dimensional RNN structure is studied on a surface code with code distances 3 and 5. The results of the study show that its viability as a decoder is limited in all tested configurations, obtaining lower accuracy than comparable models. Alterna tive methods and approaches are presented that show more promising results, while still performing under par. The study is concluded by speculating on additional alternative approaches for further study on algorithms utilizing RNNs for quantum error correction.
Quantum error correction, Variational Neural annealing, Recurrent neural networks, Surface Code.