Agnostic Data Debiasing Through a Local Sanitizer Learnt from an Adversarial Network Approach
The widespread use of automated decision processes in many areas of our society raises serious ethical issues concerning the fairness of the process and the possible resulting discriminations. In this work, we propose an approach called GANsan whose objective is to prevent the possibility of any discrimination, )direct or indirect) based on a sensitive attribute by removing the attribute itself as well as the existing correlations with the remaining attributes. Our sanitization algorithm GANsan is partially inspired by the framework of generative adversarial networks (in particular the Cycle-GANs), which offers a flexible way to learn a distribution empirically or to translate between two different distributions. In contrast to prior work, one of the strengths of our approach is that the sanitization is performed in the same space as the original data by only modifying the other attributes as little as possible and thus preserving the interpretability of the sanitized data. As a consequence, once the sanitizer is trained, it can be applied to new data, such as for instance, locally by an individual on his profile before releasing it. We explore the utility preserved by the sanitization by conducting on a dataset various experiments, which we believe represents possible real world use case of the sanitization procedure and the sanitized data. Our observations bring forward new research questions which we will briefly introduce.
I am a third year PhD student in Computer Science at Université du Québec à Montréal advised by Sébastien Gambs (UQAM) and Alain Tapp (UDEM).
My research interests lie in Machine learning, Adversarial training of models and Computer Security. Currently, I am working on Fairness, Accountability and Transparency of machine learning models, topics in which we develop new techniques to prevent the amplification of existing bias towards minorities or underrepresented population groups. In fact, as discrimination is present in our realities, we ought to formulate approach to prevent machine learning models to learn to discriminate. Similarly, we need to clearly understand the decision path of models (or their internal reasoning), to decide whether the models and their predictions are trustworthy, before taking actions based on such predictions.
I graduated from Georgia Institute of Technology (Gatech) and Université de Technologie de Troyes (UTT) with a double degree in Electrical and Computer Engineering from both schools. Prior to that, I obtained my bachelor degree in Systems Networks and Telecommunications with a speciality in Mobile Technology and Embedded systems from UTT.