I am a M.Sc. student in Data Science and Mathematics (Minor) at EPFL. Currently, I am an assistant student for Machine Learning CS-433 and working on low-rank matrix completion under supervision of Nicolas Boumal. I did my semester project on latent-space perturbations at MLO Lab, advised by Tatjana Chavdarova and Sebastian U. Stich.
Previously, I finished my Bachelor’s degree in Computer Engineering and Mathematics at Boğaziçi University, where I worked on adversarial robustness and orthogonality in neural networks, advised by İnci M. Baytaş. I also worked on normalizing flows and unsupervised discovery of control for GANs, advised by Pınar Yanardağ.
MSc in Data Science, 2020-2022
École Polytechnique Fédérale de Lausanne
BSc in Computer Engineering & Mathematics, 2015-2020
Data augmentation is a widely adopted technique for avoiding overfitting when training deep neural networks. However, this approach requires domain-specific knowledge and is often limited to a fixed set of hard-coded transformations. Recently, several works proposed to use generative models for generating semantically meaningful perturbations to train a classifier. However, because accurate encoding and decoding are critical, these methods, which use architectures that approximate the latent-variable inference, remained limited to pilot studies on small datasets.
Exploiting the exactly reversible encoder-decoder structure of normalizing flows, we perform on-manifold perturbations in the latent space to define fully unsupervised data augmentations. We demonstrate that such perturbations match the performance of advanced data augmentation techniques – reaching 96.6% test accuracy for CIFAR-10 using ResNet-18 and outperform existing methods, particularly in low data regimes – yielding 10–25% relative improvement of test accuracy from classical training. We find that our latent adversarial perturbations adaptive to the classifier throughout its training are most effective, yielding the first test accuracy improvement results on real-world datasets – CIFAR-10/100 – via latent-space perturbations.