Publications

This paper presents a novel nonlinear transform model for learning of collaboration structured, discriminative and sparse representations. The idea is to model a collaboration corrective functionality between multiple nonlinear transforms in order to reduce the uncertainty in the estimate. The focus is on the joint estimation of a data-adaptive nonlinear transforms (NTs) that take into account a collaboration component w.r.t. a discrimination target. The joint model includes minimum infor- mation loss, collaboration corrective and discriminative priors. The model parameters are learned by minimizing the empirical negative log likelihood of the model, where we propose an efficient solution by an iterative, coordinate descend algorithm. Numerical experiments validate the potential of the learning principle. The preliminary results show advantages in comparison to the state- of-the-art methods, w.r.t. the learning time, the discriminative quality and the recognition accuracy.

In this paper, we explicitly model a discriminative- ambiguous setup by two jointly learned parametric nonlinear transforms. The idea is to use one nonlinear transform for ambiguization and the other one for discrimination, and also to address a privacy-utility setup that is conditioned on am- biguization and discrimination priors, respectively, together with minimum information loss prior. The generic coupled representa- tion is composed by linear combination using the two nonlinear transforms. The model parameters are learned by minimizing the empirical log likelihood of the model, where we propose an efficient solution using block coordinate descend alternating al- gorithm. The proposed model has low computational complexity and high recognition accuracy for the authorized parties while having low recognition accuracy for the unauthorized parties. We validate the potential of the proposed approach by numerical experiments.

This paper proposes a group membership verification pro- tocol preventing the curious but honest server from recon- structing the enrolled signatures and inferring the identity of querying clients. The protocol quantizes the signatures into discrete embeddings, making reconstruction difficult. It also aggregates multiple embeddings into representative values, impeding identification. Theoretical and experimental results show the trade-off between the security and the error rates.

We investigate the privacy of two approaches to (biometric) template protection: Helper Data Systems and Sparse Ternary Coding with Ambiguation. In particular, we focus on a privacy property that is often overlooked, namely how much leakage exists about one specific binary property of one component of the feature vector. This property is e.g. the sign or an indicator that a threshold is exceeded. We provide evidence that both approaches are able to protect such sensitive binary variables, and discuss how system param- eters need to be set.

We propose a practical framework to address the problem of privacy-aware image sharing in large-scale setups. We ar- gue that, while compactness is always desired at scale, this need is more severe when trying to furthermore protect the privacy-sensitive content. We therefore encode images, such that, from one hand, representations are stored in the pub- lic domain without paying the huge cost of privacy protec- tion, but ambiguated and hence leaking no discernible content from the images, unless a combinatorially-expensive guessing mechanism is available for the attacker. From the other hand, authorized users are provided with very compact keys that can easily be kept secure. This can be used to disambiguate and reconstruct faithfully the corresponding access-granted images. We achieve this with a convolutional autoencoder of our design, where feature maps are passed independently through sparsifying transformations, providing multiple com- pact codes, each responsible for reconstructing different at- tributes of the image. The framework is tested on a large-scale database of images with public implementation available.