|제목||[학술세미나] [학과세미나] 10월 27일(금) 11시 학과세미나 안내|
▪제목 : Learning adaptive universal denoiser with deep neural networks
▪연사 : 문 태 섭 (성균관대학교, 전자전기공학부)
▪일시 : 2017년 10월 27일(금) AM 11:00 – 12:00
▪장소 : 25동 405호
In this talk, I will show how the neural networks can be adpatively trained as a universal denoiser solely with statistical “loss estimators” obtained from noisy data (and without any ground-truth labels). Such approach can be thought of a discrete version of the well-known SURE (Stein's Unbiased Risk Estimator) principle. I will then show that the performance of such adaptively trained denoiser can be boosted when combined with supervised learning, by presenting strong empirical results on two very different data sources, i.e., image and DNA sequence. Furthermore, for the case of discrete data, I will present a theoretical result on the uniform concentration of the loss estimator we use, which can not only be used for the performance analysis of the algorithm but also for the accurate selection of the hyper-parameter of the algorithm. Finally, I will conclude with some potential future research directions.
세미나 안내_171027_A4.hwp [14KB]