3. Symposium

Our third symposium will be held on Monday and Tuesday, September 8 to 9, 2014, at Magnus Haus Berlin, Kupfergraben 7, 10117 Berlin (see programme below).

The first objective of the meeting is to discuss our research, what we have achieved in phase 1 of our Priority Programme and plan to achieve within the phase 2 projects. Generally, the program mixes presentations of previous and planned research with time slots for discussion. In the latter we discuss which mid- and long-term goals people aim for, how the concluded and new projects fit into such research agendas and what concrete research steps and collaborations are most promising. The two larger topics of the symposium programme are:

  • Robotics and Reinforcement Learning
  • New ML: Auto representations, features, hyperparameters.

These are discussed on the 1st and 2nd day, respectively, with other topics mixed in. PIs, please prepare a presentation on your project – depending on your status either to summarize the research results of phase 1 and/or to introduce your project in the second phase. Each presentation should only be max 10 minutes or max 15 minutes for those with two projects, with time for wider discussion at the end of the topic blocks. For each project, please also prepare a simple poster (DIN A1 portrait) for the extended poster session on Monday evening, which will allow for more direct interaction and discussion of your research and technical aspects.

Preliminary programme

Monday, September 8, 2014

09:00 – 09:15 Reception with Coffee

09:15 – 09:45 Marc Toussaint: Welcome and information on the Priority Programme and it’s 2nd phase

09:45 – 17:00 Block 1: Robotics and Reinforcement Learning

10:45 – 11:00 Coffee break
12:00 – 13:30 Lunch break at Humboldt University Mensa Süd
15:00 – 15:30 Coffee and cake

Tamim Asfour, Rüdiger Dillmann (Karlsruhe); Helge Ritter, Robert Haschke (Bielefeld) SPP Phase 1: Robots Exploring their Bodies Autonomously (REBA)
Tamim Asfour, Helge Ritter, Robert Haschke SPP Phase 2: Robots Exploring Tools as Extensions to their Body Autonomously

Sven Behnke (Bonn) SPP Phase 1: Autonomous Learning of Bipedal Walking Stabilization
Sven Behnke, Wolfram Burgard (Freiburg) SPP Phase 2: Autonomous Active Object Learning Through Robot Manipulation

Michael Beetz (Bremen) SPP Phase 1: Autonomous Learning for Bayesian Cognitive Robotics

Marc Toussaint (Stuttgart), Kristian Kersting (Dortmund) SPP Phase 1: Relational exploration, learning and inference Foundations of autonomous learning in natural environments
Oliver Brock (Berlin), Marc Toussaint SPP Phase 2: The Physical Exploration Challenge: Robots Learning to Discover, Actuate, and Explore Degrees of Freedom in the World

Discussion: What are the core challenges in autonomously learning robots?

Klaus Obermayer (Berlin) SPP Phase 1: Value Representation in Large Factored State Spaces
Klaus Obermayer SPP Phase 2: Linking metric and symbolic levels in autonomous reinforcement learning

Frank Pasemann (Osnabrück) SPP Phase 1: Learning to Behave in the Sensorimotor Loop — An Animat Approach v

Nihat Ay, Eckehard Olbrich (Leipzig) SPP Phase 1: An information theoretic approach to autonomous learning of embodied agents
Nihat Ay SPP Phase 2:An information theoretic approach to autonomous learning of embodied agents

Eyke Hüllermeyer (Marburg), Johannes Fürnkranz (Darmstadt): Reinforcement Learning with Qualitative Feedback

Barbara Hammer (Bielefeld), Niels Pinkwart (Berlin) SPP Phase 1: Learning Feedback in Intelligent Tutoring Systems (FIT)
Barbara Hammer, Niels Pinkwart SPP Phase 2: Learning Dynamic Feedback in Intelligent Tutoring Systems

Daniel Cremers, Rudolf Triebel (München) SPP Phase 2: Efficient Active Online Learning for 3D Reconstruction and Scene Understanding

Gerhard Neumann, Jan Peters (Darmstadt) SPP Phase 2: Learning Modular Policies for Robot Motor Skills

Jan Peters, Martin Riedmiller (Freiburg) SPP Phase 2: Scalable Autonomous Reinforcement Learning – From scratch to less and less structure

Discussion: Active/Reinforcement learning for everything – and the gap to applying it in robotics.

17:00 – 18:00 Open discussion on the future of Autonomously Learning Systems (robotics, RL, AI)

18:00 Dinner buffet with Poster Session in Magnus Haus Remise

Tuesday, September 9, 2014

09:00 – 09:30 Priority Programme instruments and activities

09:30 – 12:30 Block 2: New Machine Learning – Auto representations, features, hyperparameters (see details below)

10:45 – 11:00 Coffee break

Thomas Martinetz, Erhard Barth (Lübeck) SPP Phase 1: Learning Efficient Sensing for Active Vision (Esensing)

Reinhold Häb-Umbach (Paderborn) SPP Phase 1: Sparse Coding Aroaches to Language Acquisition
Reinhold Häb-Umbach SPP Phase 2: Bayesian Learning of a Hierarchical Representation of Language from Raw Speech

Patrick van der Smagt (München) SPP Phase 1: Active exploration in the high-dimensional data of an artificial Skin

Klaus Müller (Berlin) SPP Phase 1: Theoretical concepts for co-adaptive human machine interaction with application to BCI

Gregor Schöner (Bochum) SPP Phase 1: Development as autonomous learning: Emergence of developmental stages that lead from sensori-motor behaviors to embodied cognition

Thomas Brox, Frank Hutter (Freiburg), Philipp Henning (Tübingen) SPP Phase 2: Auto-Tune: Structural Optimization of Machine Learning Frameworks for Large Datasets

Lars Schmidt-Thieme (Hildesheim) SPP Phase 2: Hyperparameter Learning Across Problems

Jörg Lücke (Oldenburg) SPP Phase 2: Autonomous and Efficiently Scalable Deep Learning

12:30 – 14:00 Lunch break at Humboldt University Mensa Süd

14:00 – 15:00 Open discussion on the future of Autonomous Machine Learning. (What are the fundamental limits to fully autonomous ML? Are there limits – will expert engineering of representations, features, system structures become obsolete, independent of the domain?)

15:00 – 15:15 Coffee and cake

15:15 – 16:30 Results, agreements and closing remarks

16:30 End of Symposium