Autonomous Learning

How a machine can autonomously learn about itself, its environment and how to control them.

Self-organization of robotic behavior

I study how coherent and complex behaviors can emerge from generic objective functions.

Welcome

Welcome to my web page about my scientific identity.

georg_portrait_2013_white
Me

I am currently building up a research group on Autonomous Learning at the Max Planck Institute for Intelligent Systems in Tübingen. I have open positions, see below.

Before joining the MPI in Tübingen, I was postdoc fellow at the IST Austria in the groups of Christoph Lampert and Gašper Tkačik after being a postdoc at the Max Planck Institute for Mathematics in the Sciences in Leipzig. I am interested in autonomous learning, that is how an embodied agent can determine what to learn, how to learn, and how to judge the learning success. I am using information theory and dynamical systems theory to formulate generic intrinsic motivations that lead to coherent behavior exploration – much like playful behavior. Furthermore, I develop methods to quantify autonomous behavior in robots and animals using information theory and manifold learning. Recently, I am also working on machine learning methods particularly suitable for internal models and learning from sequences.

Check out our website of the research network for self-organization of robot behavior for publications, software, videos etc.

Ralf Der and I published a book: The Playful Machine.

Short Curriculum Vitae

Opportunities

Are you curious how to make robots curious? Do you have a strong background in computer science or physics or mathematics or engineering? Are you interested in developing algorithms, tinkering with robots, and analyzing experiments?

We are looking for motivated Master and PhD students and also for Post-docs to join us in the quest to make robots more autonomous. Female applicants are particularly encouraged to apply. Appropriately qualified handicapped persons will be equally considered.

Our focus is on different aspects of autonomous learning with applications to robotics. We develop new algorithms and theory for making robots curious and to give them the ability to autonomously learn how to act. We apply our methods to real robots, in particular, soft and tendon driven systems which we partly build in the lab. Research topics include and combine reinforcement learning, information theory, self-organization, dynamical systems, deep learning and also some mechanical engineering and 3D printing. We largely work with and Python, C++ and Mathematica.

Team: You will be part of a small young team embedded in the prestigious Max Planck Institute for Intelligent Systems with its core focus on Perception - Learning - Action and many outstanding scientists next door, check out the Tübingen campus.

Time and funding: The positions are available from May 2017. The appointment for PhD students will be for three years and for Post-docs for one and can then be extended if needed. The positions are fully funded with competitive salaries and full social security (health insurance etc.).

Requirements: Candidates should have an education in computer science, physics, mathematics, electrical engineering or any related field. Required are strong mathematical and programming skills as well as the ability to communicate and work well in a team.

Deadline: Official deadline is March 18., but later applications might still be considered. Interviews will take place in the beginning of April.

Instructions: Prospective applicants can apply by sending an email to <georg dot martius at tuebingen.mpg.de>. The subject should contain "Application AL42", your name and the position you are applying to. (e.g. Application AL42 PhD Max Smith). Include the following material in pdf format (filename(s) firstname_lastname_{CV|Letter|all|...}  and keep things short):

  1. a letter of motivation, (pdf or plain text in email)
  2. CV,
  3. a copy of the current academic transcripts (for students), or track record
  4. contact details of at least 2 references including their academic relation to you, e.g. thesis supervisors.job-application

Applications have to be in English and the letter should include answers to the following questions:

  1. What are your abilities in regard to mathematical pen-and-paper calculations, modeling and writing code?
  2. What are your abilities/experience with robotics and low-level hardware?
  3. Why would you like to work in robotics? What sort of project you think would be ideal for you?
  4. How comfortable do you feel writing in English and do you enjoy it?

Note that not all skills are required, I just need to know to build a good team. Inquiries can be sent to the same address, please include the code "Application AL42" in any case.

Looking forward to your excellent application.

 

Publications

 

Complete list of my publication, or take a look at my google scholar page.

 

 

Activities

HAL 10: Haskell Workshop in Leipzig

HAL 10: Haskell Workshop in Leipzig

Haskell is a modern functional programming language fostering the rapid development robust and correct software.

Since 10 years we organize in Leipzig and Halle (Germany) an annual workshop called HAL on theory and application of Haskell. Big thanks to Johannes Waldmann for his major efforts in Organization.

See webpage for more details.

4. und 5. Dezember 2015, HTWK Leipzig

 

Workshop: Information Theory in Artificial Life on ECAL 2015

Workshop: Information Theory in Artificial Life on ECAL 2015

Information Theory in Artificial Life

Monday 20th July 2015, 10:00 – 13:00

In the workshop “Information Theory in Artificial Life” we will discuss how Information Theory can be used to generate, motivate, understand and quantify the behaviour and other processes in artificial agents and life-like systems. Information Theory provides a language to express the required concepts and quantities in a general way, allowing to transfer them between different domains. For further details and information on how to submit and participate see our website. If you have any questions please send us an email at itialife@gmail.com.

Organised by:  Georg MartiusChristoph SalgeKeyan Ghazi-Zahedi, and Daniel Polani

Workshop: Information Theoretic Incentives for Artificial Life on ALIFE 2014

Workshop: Information Theoretic Incentives for Artificial Life on ALIFE 2014

Satellite Workshop of ALife 2014 (14th International Conference on the Synthesis and Simulation of Living Systems)

New York, 30th of July 2014

Workshop: Conceptual and Mathematical Foundations of Embodied Intelligence

Workshop: Conceptual and Mathematical Foundations of Embodied Intelligence

February 27 – March 01, 2013

Website

Abstract: In the last few decades, an overwhelming number of case studies produced the evidence that intelligent behavior of naturally evolved agents efficiently involves the embodiment as part of the underlying control process. Nowadays, there is no question that the exploration and exploitation of the embodiment represent important mechanisms of cognition. The shift from the classical view to the modern embodied view, also referred to as the cognitive turn, not only framed a novel way of thinking about intelligence but also identified a number of fundamental principles that intelligent systems obey. Well known examples are the principle of cheap design, morphological computation, and information self-structuring. Although there is general consensus on the intuitive meaning of such principles, the field of embodied intelligence currently lacks a formal theory. We think that the mathematical foundations of the core concepts have to be advanced and unified, in order to be able to realize and better understand cognitive systems that exploit their embodiment in an autonomous and completely intrinsic way. Information theory, dynamical systems theory, and information geometry already turned out to be useful in this regard. However, there is much more, and also much more to do.

Summarizing, the goal of the workshop is to identify the core concepts and to advance the theoretical foundations of embodied intelligence.

Invited speakers:

  • Randall D. Beer (Indiana University, USA): Information and dynamics in brain-body-environment systems
  • Paul Bourgine (École Polytechnique, France): Paradigms and models of embodied intelligence
  • Karl Friston (University College London, United Kingdom): Embodied inference and free energy
  • David Krakauer (University of Wisconsin-Madison, USA): Agents and their artefacts: a natural history of ex-bodiment
  • Thomas Metzinger (Johannes Gutenberg-Universität Mainz, Germany): Body-representation and self-consciousness: from embodiment to minimal phenomenal selfhood
  • Kevin O’Regan (Centre National de Recherche Scientifique, France): A theoretical basis for how artificial or biological agents can construct the basic notion of space
  • Frank Pasemann (Universität Osnabrück, Germany): Neurodynamics in the sensorimotor loop
  • Daniel Polani (University of Hertfordshire, United Kingdom): The role of information in formation of cognitive organization
  • Helge Ritter (Universität Bielefeld, Germany): Manual intelligence and embodiment
  • Jürgen Schmidhuber (Istituto Dalle Molle di Studi sull’Intelligenza Artificiale, Switzerland, and Technische Universität München, Germany): Optimal AI – neural network ReNNaissance – theory of fun
  • Naftali Tishby (The Hebrew University, Israel): Information flow in sensation & action and the emergence of [reverse] hierarchies

Organizers: Nihat AyRalf DerKeyan Ghazi-ZahediGeorg Martius

 

Videos