Fig. 4: Inference time.
From: Gesture recognition with Brownian reservoir computing using geometrically confined skyrmion dynamics
Comparison of the inference time per gesture for the linear readout layer of the reservoir (LIN SVM on 5 MTJ Reservoir) versus a non-linear radial basis support vector machine (RBF SVM) operating directly on the Range-Doppler (RD) data. The best- and worst-case scenarios demonstrate inference times based on the number of Kerr images used. In the worst-case scenario, all Kerr frames are used, resulting in a more complex readout layer. Each benchmark is run 1000 times for each data point.
