Sven Gowal (DeepMind)
After receiving his Master’s degree from the Swiss Federal Institute of Technology in Lausanne (EPFL), in October 2007, Sven Gowal joined the Distributed Intelligent Systems and Algorithms Laboratory (DISAL) in May 2008. He started his PhD thesis under the supervision of Prof. Alcherio Martinoli in October 2008 and is studying graph-theoretic methods to control large groups of robots.
Short Abstract: Since the work by Madry et al. (2017), on CIFAR-10, the accuracy of robust models against $\ell_\infty$-norm bounded adversarial perturbations of size 8/255 has steadily increased from 44% to 66%. In this presentation, we take a look at the most recent developments and try to make an educated guess on where we see the field progressing. We elaborate on different hypotheses in the hope of establishing future research directions.