Learning Semi-Supervised Anonymized Representations by Mutual Information

Clément Feutry (1) , Pablo Piantanida (2, 1) , Pierre Duhamel (1)

1 L2S - Laboratoire des signaux et systèmes
2 MILA - Montreal Institute for Learning Algorithms [Montréal]

This paper addresses the problem of removing from a set of data (here images) a given private information, while still allowing other utilities on the processed data. This is obtained by training concurrently a GAN-like discriminator and an autoencoder. The optimization of the resulting structure involves a novel surrogate of the misclassification probability of the information to remove. Several examples are given, demonstrating that a good level of privacy can be obtained on images at the cost of the introduction of very small artifacts.

Domains: Computer Science [cs]; Information Theory [cs.IT]; Computer Science [cs]; Networking and Internet Architecture [cs.NI]; Mathematics [math]; Information Theory [math.IT]; Mathematics [math]; Statistics [math.ST]

Clément Feutry, Pablo Piantanida, Pierre Duhamel. Learning Semi-Supervised Anonymized Representations by Mutual Information. ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), May 2020, Barcelona, Spain, France. pp.3467-3471, ⟨10.1109/ICASSP40776.2020.9053379⟩. ⟨hal-03351090v2⟩