Biography

Georg Martius is a research group leader at the Max Planck Institute for  Intelligent Systems in T\”ubingen, Germany since 2017. His background is in computer science, however, his interest is on developmental robotics, machine learning, information theory, artificial life and neuroscience. He was holding postdoc positions at IST Austria and the MPI for Mathematics in the Sciences in Leipzig. Currently, he is investigating principles and algorithms for autonomous learning systems. He authored a monograph called “The Playful Machine” published in Springer 2012.

 

Abstract

Control what you can

I will present present a hierarchical reinforcement learning agent that is intrinsically motivated to learn how to control its observation space in the  fastest possible manner by optimizing learning progress. The agent mimicks to some extend the devleopment process how a baby learns. Our agent learns what can be controlled, how to allocate time and attention, and the relations between objects in the environment. In a nutshell, our work combines several task-level planning ideas (e.g., backtracking search on task graph, probabilistic road-maps, allocation of search efforts, etc.), in the form of structured prior, with deep RL control and relational reasoning to learn from scratch.