avg_rank_experiments.pdf

The average risk after training a 6-qubit QNN for randomly generated target unitaries. For this experiment, data of varying Schmidt rank with average ranks r ∈ {1,2,4,64} and different input sizes t ∈ {1,2,4,8,16,32,64} was used. The markers give the experimentally computed average risk for each pair of Schmidt rank and training set size. The lower bounds according according to the QNFL theorem are shown as solid lines.