Fig. 4: Parallelized convolution of gait signals from patients with Parkinson’s disease.
From: Partial coherence enhances parallelized photonic computing
a, Schematic of the computing system using the 3 × 3 photonic memory tensor core. The partially coherent lights in use are 0.8-nm-bandwidth C34 and C33. b, Convolution results of a typical gait signal from three different kernels when using a partially coherent system in (i) and a coherent system in (ii). All of the convolutions are performed once. The error bands represent the standard deviation of convolution results from 50 gait signals generated by the same patient, showing the variation of gait signals from this patient. c, Convolution accuracy of using a partially coherent system in (i) and a coherent system in (ii). A total of 43,500 pairs of expected and measured results are compared in each system. The insets show the Gaussian distribution of normalized errors. d, CNN architecture. e, Comparison of CNN classification results. a.u., arbitrary units; PD, photodetector.
