Fig. 8: Performance of the Reservoir, when different amount of Kerr video frames is used for the evaluation. | Nature Communications

Fig. 8: Performance of the Reservoir, when different amount of Kerr video frames is used for the evaluation.

From: Gesture recognition with Brownian reservoir computing using geometrically confined skyrmion dynamics

Fig. 8

The plots show the pairwise gesture comparison 0-push vs 1-swipe left. Error is the standard error. We see that the performance on the validation set peaks at specific amount of input (in this case 0.708 for 87 data points), while above we enter an overfitting regime of the linear SVM. For the plots in Fig. 3 in the main text, we plot the best performance we find in the validation set. The best MTJ position for varying input size is also indicated in Fig. 2d) in the main text.

Back to article page