Presentation Information
[23a-P09-8]Optimization of Magneto-Optical Diffraction Deep Neural Networks (2)
〇(M1)Reo Akagawa1, Hotaka Sakaguchi1, Hirofumi Nonaka2, Hiroyuki Awano3, Fatima Zahara Chafi1, Takayuki Ishibashi1 (1.Nagaoka Univ. Tech., 2.Aichi Inst. Tech., 3.Toyota Tech. Inst.)
Keywords:
Deep Neural Network,optics,Magneto optical effect
In complex models of Deep Neural Networks, increased processing speed and power consumption have become issues. To address these, we have proposed an optical diffraction-based Deep Neural Network utilizing the magneto-optical effect, termed MO-D2NN. However, MO-D2NN has many parameters that need consideration. In this study, our goal was to investigate the impact of light intensity distribution on computational accuracy when the polarization plane's rotation angle is used as the output signal for online learning. Accordingly, we evaluated the case using input data with significantly different light intensity distributions.