DeepMind Research Scientists Presentation Series at Politehnica University of Timisoara

Abstract: The first part of the talk represents a discussion on the continual learning problem for neural networks and its importance for the field. Then it concentrates on Progress & Compress (https://arxiv.org/abs/1805.06370), a recent framework for continual learning published at ICML 2018, and discuss some of the open issues on this topic.

Short Bio: Razvan obtained his Ph.D. at Universite de Montreal, working with Yoshua Bengio on optimization, recurrent models and deep learning in general. His interests range from topics like optimization, neural networks to deep reinforcement learning, continual learning and structured neural networks models.

Abstract: The presentation covers our recent work on designing efficient video networks (https://arxiv.org/pdf/1806.03863.pdf ECCV2018) where we propose a new framework for deep video processing that leverages operation pipelining and multi-rate clocks within deep neural networks.

Short Bio: Viorica obtained her PhD from University of Toulouse in statistical image processing. She was a postdoc with Ecole Polytechnique Paris and University of Cambridge working on 3D and video processing, prior to joining DeepMind. She works mainly on Computer Vision problems.

15 min. presentation of the first edition of the TMLSS - Transylvanian Machine Learning Summer School and Q&A session.