Deep Neural Network Based Optical Monitor Providing Self-Confidence as Auxiliary Output
Takahito Tanimura, Tomoyuki Kato, Shigeki Watanabe, Takeshi Hoshida · 2018
We present a deep-learning-based optical monitor that simultaneously outputs an optical signal-to-noise ratio and its confidence level using a dropout technique at inference time. Its behaviour is analysed with partially missing and limited number of records in training dataset.