Towards RL that scales - Victor Campos - UPC TelecomBCN Barcelona 2020

Course site: https://ift.tt/34VAJ2w Towards RL that scales: Autonomous acquisition and transfer of knowledge ABSTRACT Designing agents that acquire knowledge autonomously and use it to solve new tasks efficiently is an important challenge in reinforcement learning (RL). Unsupervised learning provides a useful paradigm for autonomous acquisition of task-agnostic knowledge. In supervised settings, representations discovered through unsupervised pre-training offer important benefits when transferred to downstream tasks. In this talk, we discuss whether such techniques are well suited for RL. While reviewing recently proposed approaches for unsupervised pre-training of RL agents, we will gain insight on the key aspects that enable autonomous acquisition and efficient transfer of knowledge in our agents. BIO Víctor Campos holds a BsC and a MsC degrees in Electrical Engineering from Universitat Politècnica de Catalunya. He is currently pursuing his PhD on the intersection between Deep Learning and High Performance Computing at the Barcelona Supercomputing Center, supported by Obra Social "la Caixa" through La Caixa-Severo Ochoa International Doctoral Fellowship program. He has done internships at DFKI (2016), Columbia University (2017), Salesforce Research (2019), and Deepmind (2020). His research interests focus on scaling up deep learning and reinforcement learning methods to leverage compute and data.