Ruben Glatt
Reinforcement Learning, Deep Learning, Transfer Learning
Supervisor: Prof. Anna Helena Reali Costa

Location: São Paulo, Brazil

Contact: ruben.glatt [at]

Profile on Research Gate    Profile on Google Scholar    Profile on Lattes    Profile on LinkedIn   Profile on Google+

Rubens PhD project aims to combine the successful approaches of Deep Learning, Reinforcement Learning and Transfer Learning to improve the generalization of knowledge through abstract representation and autonomous similarity determination in the area of single agent systems and robotics.
Inspired by the recent developments in Deep Learning mainly through the works of Geoffrey Hinton, Yoshua Bengio and Yann LeCun, Rubens research initially focused on exploiting already gathered knowledge in the field. Impressed by the results of Google’s Deep Mind group in the area of Deep Reinforcement Learning for intelligent game playing agents, he decided to make this the main content of his PhD project. The key question of this proposal is the following: how can we exploit state-of- the-art ML techniques such as DL to provide a robust framework for RL that is able to generalize and transfer knowledge to several challenging new tasks within a related domain?
The goal is to develop an extension of the Deep Q-Network algorithm as proposed in this article published in Nature, which is able to provide a generalization of a learned model and transfer the gathered knowledge to a new task from a related domain using the same model. Using this approach the intend is to speed up the time to convergence for follow-up tasks, provide a more general model and a better approximation for unknown states. The challenge is to preserve the critical information necessary to solve a concrete task while generalizing enough to be able to transfer the knowledge to other tasks.
His project proposal Improving Deep Reinforcement Learning through Knowledge Transfer has received the GOOGLE Research Award Latin-America 2015 as one of only twelve winning projects. 
Ruben has a German engineering Diplom degree in Mechatronics, Dipl.Ing. Mechatronik, from the Karlsruhe Institute of Technology (KIT) and a Brazilian engineering degree in Mechanics, Mestre em Engenharia Mecânica, from the Universidade Estadual Paulista Júlio de Mesquita Filho (UNESP) in São Paulo. During his German degree program his main topics involved Sensors, Robotics and Integrated Production Planing, while focusing on an application for a Deep Learning system for the recognition of sign gestures in his Brazilian master.
Apart from his education Ruben has over 10 years of experience in data center environments, which he acquired before and during his studies in full and part time working in Germany and the United States. His experiences include administration and development of data center control and monitoring systems of several data centers in Germany and the USA, supporting data center infrastructure optimization based on data center monitoring analysis and evaluation of new technologies, supervision of the first United Internet datacenter co-location in the US and support and training of staff in the initial phase of building the first own datacenter of the United Internet group in the US.
For more details, please refer to his profile on linkedIn.
Things to know about Ruben:
• Now second behind AI and ML is Rubens passion for Squash (Former Professional Player, Highest World Ranking: 206)
• He used to play a lot of and is still a great fan of soccer (Bayern München, KSC)
• Gets stuck a lot on Netflix (SciFi, Fantasy) and TED talks…
• Frequent online learner (Coursera, edX, etc.)