The paper entitled "Supervised Joint Nonlinear Transform Learning with Discriminative-Ambiguous Prior for Generic Privacy-Preserved Features" has been accepted for lecture presentation at 53rd Annual Conference on Information Systems & Sciences (CISS 2019), to be held Johns Hopkins University in Baltimore, Maryland. 




Fig 1. Extracting the discriminative representation and ambiguous representation from the corresponding learned nonlinear transforms, and obtaining the final privacy-protected representation.



Fig 2. Visualizing three classes of data (i) in the original domain and (ii) in the transform domain.