Magneto- optical diffractive deep neural network

We propose a magneto-optical diffractive deep neural network (MO-(DNN)-N-2). We simulated several MO-D(2)NNs, each of which consists of five hidden layers made of a magnetic material that contains 100 x 100 magnetic domains with a domain width of 1 mu m and an interlayer distance of 0.7 mm. The networks demonstrate a classification accuracy of > 90% for the MNIST dataset when light intensity is used as the classification measure. Moreover, an accuracy of > 80% is obtained even for a small Faraday rotation angle of p/100 rad when the angle of polarization is used as the classification measure. The MO-(DNN)-N-2 allows the hidden layers to be rewritten, which is not possible with previous implementations of D(2)NNs.

Automated summary

We propose a new network architecture called magneto-optical diffractive deep neural network (MO-(DNN)-N-2), which allows rewriting of hidden layers. Through simulations and testing, we found that this network achieves high classification accuracy on the MNIST dataset using different classification measures, such as light intensity and polarization angle.