The Good, the Bad and the Ugly in Deep Learning: a personal view of current state, open problems and shortcomings of the field.
I will discuss through a highly personal and biased perspective a number of recent topics in deep learning. I will illustrate with examples the existing gap between our current theoretical understanding and the latest empirical results. I will then argue why this gap is both a curse and a blessing, and what are its potential risks and opportunities in the context of both academic and industrial research. I will conclude with some open questions and ongoing research projects with my collaborators, involving algorithmic discovery and complexity.
Joan Bruna graduated from ENS (France) in Applied Mathematics after an undergrad at UPC Barcelona (Cfis). He then spent 5 years in a startup as a Research engineer, working in realtime video processing algorithms. He then went back to academia and obtained his PhD in Applied Mathematics at Ecole Polytechnique (France) in 2013, under the supervision of Stephane Mallat. After a postdoctoral appointment at the Courant Institute (NYU, New York) he became a postdoctoral fellow at FAIR (Facebook AI Research), New York. Since 2015 he is Assistant Professor at UC Berkeley, working in various areas of Machine Learning, including mathematical aspects of Deep Learning.