[ITuPJ] Optical Information

[ITuPJ-01] Dependence of activation function on image recognition accuracy in self-referential holographic deep neural network

*Rio Tomioka¹, Masanori Takabayashi¹ (1. Kyushu Inst. of Tech. (Japan))

[Presentation Style] Onsite

We numerically compared the image recognition properties of self-referential holographic deep neural network (SR-HDNN) with various activation functions. The image recognition rate was improved to 87.1% by applying intensity-ReLU function to SR-HDNN.