SLIDE 44 Sebastian Raschka Big Data Madison, Aug 2019 44
Gender Privacy: An Ensemble of Semi Adversarial Networks for Confounding Arbitrary Gender Classifiers
Vahid Mirjalili, Sebastian Raschka, and Arun Ross (2018) Gender Privacy: An Ensemble of Semi Adversarial Networks for Confounding Arbitrary Gender Classifiers. 9th IEEE International Conference on Biometrics: Theory, Applications, and Systems (BTAS 2018)
Hypothesis space of gender classifiers
G2 G1 G3 G4 G5 G2 G1 G3 G4 G5 Non-diverse: Ensemble SAN cannot generalize Diverse: Ensemble SAN can generalize
Figure 1: Diversity in an ensemble SAN can be enhanced through its auxiliary gender classifiers (see Figure 2). When the auxiliary gender classifiers lack diversity, ensemble SAN cannot generalize well to arbitrary gender classifiers.
3.2. Diversity in Autoencoder Ensembles
Training Phase Evaluation Phase
S1 S2 St
X Y1 Y2 Yt
S2
X Y
S1
X Y
St
X Y
Figure 4: Face prototypes computed for each group of at- tribute labels. The abbreviations at the bottom of each im- age refer to the prototype attribute-classes, where Y=young, O=old, M=male, F=female, W=white, B=black.
Improvements to construct a more diverse set of SAN models for better generalizability via ensembling