Repository navigation
Home
This circuit consists of a transversal CNOT, in Steane's error correction scheme, using Steane's 7 qubit code.
2017-10-27-01-29-02 A feedforward network with a single hidden layer, mixing syndrome X1 and X2 data, the output layer, mixes syndromes X3 and X4 with the output of the hidden layer.
Also includes: 2017-10-27-16-21-51 , 2017-10-27-20-02-36 and 2017-10-27-21-19-53 .
2017-10-29-05-15-53 With 5 hidden layers, 4 for each of the EC blocks and a middle layer for the CNOT. There is no improvement over the single layer layout.
2017-11-02-00-16-58 A single LSTM cell receiving a sequence of size two. This is trained on X data but tested both on the X and the Z data. The input sequence is syndromes 12 and 34. The first output layer is ignored and the second output is softmax optimized. The RNN improves the results of the feedforward network. Zoomed in to the e-04 region we have 2017-11-02-18-43-37 .
2017-11-02-04-21-48 The 'lab' experiment. In this experiment the training data only consists of the indication as to whether after the candidate recovery, there is a persistent logical fault in each of the control and data logical qubits. The results are worse than the look up table, but we observed some learning. In absence of a noise model this method can work better than simulation with an unrealistic noise model - look up table analysis.
In 2017-11-03-04-01-54 , and 2017-11-03-23-51-20 the RNN is trained both on the X syndrome-error data and the Z syndrome-error data with the reuse=True flag. Turing the flag off results 2017-11-04-14-33-14.
The run 2017-11-04-14-33-14 is our best result so far before hyperparameter turning.