Figure 2.

10-fold cross-validation using the proposed method for the MNIST datasets and various training set sizes. Each cross-validation run was repeated over 100 random splits, resulting in P = 1000 total problems. The effective problem size represents the number of problems that can be solved by the direct method in the same amount of time as the P problems for the proposed method. Here, the ratio of 1000 to the effective problem size gives the computational speedup, which varies between 30× to 100×.