SPP 1527: Autonomous Learning
Biology
Mathematics
Medicine
Physics
Social and Behavioural Sciences
Final Report Abstract
There could not be a more timely topic in AI as Autonomous Learning. During the term of our priority programme AI has developed massively and disruptively, with Deep Learning and Deep RL emerging and now dominating the field. The big-data ML problem of extracting a predictor when enough data is available made enormous progress. But the core scientific challenges defined by our initiative are still pressing: • Let AI be more autonomous about its choice of representation, model, structural priors, hyperparameters, so that it can generalize truly strongly from the available data. • Let AI be more autonomous in collecting and generating its own data. • Realize strongly generalizing AI on real-world autonomous systems, to ensure that AI can also capture, model and exploit the particular structure of the one physical world we live in. The goal of the priority programme was to foster research to tackle these scientific challenges, to support the qualification of young researchers in these research areas, and to support our PIs and young researchers to strengthen their impact and networking with the growing local and international communities working on these research areas. The scientific and structural impact of the second phase of our priority programme was substantial. With an estimated 129 peer-reviewed papers (51 journal, 78 conference), 12 PhD students graduating within the programme, and 6 expected to graduate soon, and several PIs winning full professorships during, the second phase of the programme was highly successful. The SPP greatly supported collaborations, in particular international collaborations based on the NSF-DFG Collaborative Research Programme, and thereby substantially supported our PIs and young researchers to establish themselves in the growing international communities on the topic. Today, Autonomous Learning is not anymore a small niche topic with a small local research network, instead its research agenda became central to modern international AI research. Our PIs and the research agenda of autonomous learning now play a major role in the top conferences in the field.
Publications
- Construction of approximation spaces for reinforcement learning. Journal of Machine Learning Research, 2013
Böhmer, Grünew älder, Shen, Musial, Obermayer
- Hierarchical System for Word Discovery Exploiting DTW-Based Initialization. In Proc. IEEE Automatic Speech Recognition and Understanding Workshop (ASRU 2013)
Walter, Korthals, Haeb-Umbach, Raj
(See online at https://doi.org/10.1109/ASRU.2013.6707761) - Example-based feedback provision using structured solution spaces. International Journal of Learning Technology, 2014
Gross, Mokbel, Paaßen, Hammer, Pinkwart
(See online at https://doi.org/10.1504/IJLT.2014.065752) - A Theory of Cheap Control in Embodied Systems. PLoS Computational Biology (2015) 11(9)
Montúfar, Ghazi-Zahedi, Ay
(See online at https://doi.org/10.1371/journal.pcbi.1004427) - Autonomous learning of state representations for control: An emerging field aims to autonomously learn state representations for reinforcement learning agents from their real-world sensor observations. KI - Künstliche Intelligenz, 2015
Böhmer, Springenberg, Boedecker, Riedmiller, Obermayer
(See online at https://doi.org/10.1007/s13218-015-0356-1) - Efficient and robust automated machine learning International Conference on Advances in Neural Information Processing Systems, 2015
Feurer, Klein, Eggensperger, Springenberg, Blum, Hutter
- Embed to control: A locally linear latent dynamics model for control from raw images. In AdvancesNeural Information Processing Systems (NIPS), 2015
Watter, Springenberg, Boedecker, Riedmiller
- Bayesian optimization with robust Bayesian neural networks. International Conference on Advances in Neural Information Processing Systems, 2016
Springenberg, Klein, Falkner, Hutter
- Evaluating Morphological Computation in Biomechanical Muscle Models. Frontiers in Robotics and AI (2016)
Ghazi-Zahedi, Haeufle, Montúfar, Schmitt, Ay
(See online at https://doi.org/10.3389/frobt.2016.00042) - Deep reinforcement learning with successor features for navigation across similar environments. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2017
Zhang, Springenberg, Boedecker, Burgard
(See online at https://doi.org/10.1109/IROS.2017.8206049) - Hidden Markov Model Variational Autoencoder for Acoustic Unit Discovery. In Proc. Interspeech 2017
Ebbers, Heymann, Drude, Glarner, Haeb-Umbach, Raj
(See online at https://doi.org/10.21437/Interspeech.2017-1160) - Local Bayesian Optimization of Motor Skills International Conference on Machine Learning (ICML), 2017
Akrour, Sorokin, Peters, Neumann
- Opening a lockbox through physical exploration. In Proceedings of the IEEE International Conference on Humanoid Robots (Humanoids), 2017
Baum, Bernstein, Martín-Martín, Höfer, Kulick, Toussaint, Kacelnik, Brock
(See online at https://doi.org/10.1109/HUMANOIDS.2017.8246913) - Active tactile exploration based on cost-aware information gain maximization. International Journal of Humanoid Robotics, 2018
Ottenhaus, Kaul, Vahrenkamp, Asfour
(See online at https://doi.org/10.1142/S0219843618500159) - Can clustering scale sublinearly with its clusters? A variational EM acceleration of GMMs and k-means. International Conference on Artificial Intelligence and Statistics (AISTATS), 2018
Forster, Lücke
- Estimating an articulated tool’s kinematics via visuo-tactile based robotic interactive manipulation. In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2018
Li, Uckermann, Haschke, Ritter
(See online at https://doi.org/10.1109/IROS.2018.8594295) - Learning manipulation skills from a single demonstration. The International Journal of Robotics Research, 2018
Englert, Toussaint
(See online at https://doi.org/10.1177%2F0278364917743795) - Model-Free Trajectory-based Policy Optimization with Monotonic Improvement. Journal of Machine Learning Research, 2018
Akrour, Abdolmaleki, Abdulsamad, Peters, Neumann
(See online at https://doi.org/10.15496/publikation-39138) - Neural Simpletrons - Learning in the Limit of Few Labels with Directed Generative Networks. Neural Computation, 2018
Forster, Sheikh, Lücke
(See online at https://doi.org/10.1162/neco_a_01100) - The continuous hint factory - providing hints in vast and sparsely populated edit distance spaces. Journal of Educational Datamining, 2018
Paaßen, Hammer, Price, Barnes, Gross, Pinkwart
(See online at https://doi.org/10.5281/zenodo.3554698)
