An information theoretic approach to autonomous learning of embodied agents

• Applicant: Professor Dr. Nihat Ay
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing ; Human Cognitive and Systems Neuroscience ; Mathematics ; Statistical Physics, Nonlinear Dynamics, Complex Systems, Soft and Fluid Matter, Biological Physics
• Term: from 2011 to 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 200306544

• Final Report Year: 2019

Final Report Abstract

In summary, the project has achieved its foremost goal to develop geometric ways of designing low-dimesional learning systems. We have a detailed understanding of the geometry of learning systems, based on corresponding embodiment constraints. These systems are modelled in terms of low-dimensional manifolds, which formalises the notion of cheap control within the field of embodied intelligence. Furthermore, we have developed a set on natural measures of morphological computation which provides means for the exploration of body-specific behaviours. Exploration, one of the core themes of the project, was originally intended to be coupled with the above-mentioned low-dimensional manifolds. However, this coupling could not be realised, partly due to the partial funding of the project. On the other hand, the project has initiated a number of research directions, including the subject of exploration, which are pursued by previous members of the project who advanced their careers through the project, and who are now conducting their own independent work as outcome of the project.

Publications

• A Theory of Cheap Control in Embodied Systems. PLoS Computational Biology (2015) 11(9): e1004427
  G. Montúfar, K. Ghazi-Zahedi, N. Ay
  (See online at https://doi.org/10.1371/journal.pcbi.1004427)
• Geometric Design Principles for Brains of Embodied Agents. KI - Künstliche Intelligenz (2015) 29: 389–399
  N. Ay
  (See online at https://doi.org/10.1007/s13218-015-0382-z)
• Geometry and Expressive Power of Conditional Restricted Boltzmann Machines. Journal of Machine Learning Research 16 (2015) 2405–2436
  G. Montúfar, N. Ay, K. Ghazi-Zahedi
  
• Hierarchical Quantification of Synergy in Channels. Frontiers in Robotics and AI (2015)
  P. Perrone, N. Ay
  (See online at https://doi.org/10.3389/frobt.2015.00035)
• Information Geometry on Complexity and Stochastic Interaction. Entropy (2015) 17(4): 2432–2458
  N. Ay
  (See online at https://doi.org/10.3390/e17042432)
• The Umwelt of an Embodied Agent – A Measure-Theoretic Definition. Theory in Biosciences (2015) 134: 105–116
  N. Ay, W. Löhr
  (See online at https://doi.org/10.1007/s12064-015-0217-3)
• Evaluating Morphological Computation in Biomechanical Muscle Models. Frontiers in Robotics and AI (2016)
  K. Ghazi-Zahedi, D. Haeufle, G. Montúfar, S. Schmitt, N. Ay
  (See online at https://doi.org/10.3389/frobt.2016.00042)
• Comparing Information-Theoretic Measures of Complexity in Boltzmann Machines. Entropy (2017) 19(7): 310
  M.S. Kanwal, J.A. Grochow, N. Ay
  (See online at https://doi.org/10.3390/e19070310)
• Information Geometry. Ergebnisse der Mathematik und ihrer Grenzgebiete, Springer 2017
  N. Ay, J. Jost, H. V. Lê, and L. Schwachhöfer
  (See online at https://doi.org/10.1007/978-3-319-56478-4)
• Morphological Computation: Synergy of Body and Brain. Entropy (2017) 19(9): 456
  K. Ghazi-Zahedi, C. Langer, N. Ay
  (See online at https://doi.org/10.3390/e19090456)
• Morphological Intelligence – Measuring the Body’s Contribution to Intelligence. Cham: Springer 2019. ISBN 978-3-030-20621-5
  Keyan Ghazi-Zahedi
  (See online at https://doi.org/10.1007/978-3-030-20621-5)