Navigation and service

Jülich Quantum Computing Seminar:

Lukas Boedeker, PGI-2/IQI

Online Talk

30 Nov 2021 14:00

Neural network decoders for color codests


Most active error correction schemes need a classical algorithm, the so-called decoder, to interpret measurement information in order to identify the errors that occurred. Machine learning techniques seem to be suited to construct an optimizable decoder that is able to adapt to realistic noise models, which are not precisely known in general, as well as to experimental data [1, 2]. In this talk I will present a decoder for color codes, based on a recurrent neural network that can process a variable number of measurement rounds. A ”flag”- based stabilizer readout scheme [3] is used to make fault tolerant error correction possible for circuit level noise models, and we show by a scaling argument that the neural network discovers a fault tolerant decoding strategy for the distanced= 3 Steane code. The neural network decoder is further trained according to a realistic noise model which mimics the ion trap noise. In this scenario it can be shown that the neural network decoder can outperform a standard look-up table decoder.

[1] P. Baireuther et al., “Neural network decoder for topological color codes with circuit level noise”, New Journal of Physics 21, 013003 (2019).

[2] S. Varsamopoulos, B. Criger, and K. Bertels, “Decoding small surface codes with feedforward neural networks”, Quantum Science and Technology 3, 015004 (2017).

[3] C. Chamberland and M. E. Beverland, “Flag fault-tolerant error correction with arbitrary distance codes”, Quantum2, 53 (2018).

Access details

The "Jülich Quantum Computing Seminar" will take place every fortnight on Tuesdays at 14:00 CET via video conference and is intended to promote cooperation between the working groups conducting research in the field of quantum computing and quantum information at various institutes at Forschungszentrum Jülich.

Here is the access data for all the seminars:

Zoom Link:

Meeting ID: 813 1452 6642
Passcode: 01


Daniel Zeuch
Phone: +49 2461 61-6460