Presentation Information
[8p-N205-1]Efficient Training Method for Self-Referential Holographic Data Storage
with Built-in Denoising Function by Deep Learning
〇Yuta Eto1, Rio Tomioka1, Masanori Takabayashi1,2 (1.Kyushu Inst. of Tech., 2.Neumorph Center, Kyushu Inst. of Tech.)
Keywords:
holographic data storage,deep neural network,self-referential holography
We propose an efficient training method for self-referential holographic data storage with built-in denoising function. In the proposed approach, the training is performed using only a portion of the data page, and the trained parameters are applied to the entire page, enabling low-cost training independent of page size. Specifically, denoising simulations were conducted on data pages ranging from 32×32 to 128×128 pixels using training results obtained from 16×16 pixel patches. In all cases, an improvement in signal-to-noise ratio (SNR) was observed.