jts Jost Tobias Springenberg | about | research | publications login
Jost Tobias Springenberg
PhD student: University of Freiburg in Computer Science.
Advisor: Martin Riedmiller.
Interests/Keywords:Deep learning; Unsupervised learning; Invariant representations; Perception; Vision; Reinforcement learning.

E-mail:
Office: building 79 room 00010
"As great scientists have said and as all children know, it is above all by the imagination that we achieve perception [...]" Ursula K. LeGuin

About me:

I am a PhD student in the machine learning lab of the University of Freiburg, supervised by Martin Riedmiller. Prior to dedicating most of my time to research I studied Cognitive Science at the University of Osnabrueck. During my time in Osnabrueck I worked on improving a team of autonomous soccer playing agents -- the 'Brainstormers' -- and developed an interest in neural networks and statistical modelling. I then went to obtain a Master's degree in computer science from the University of Freiburg, focusing on representation learning with deep neural networks for computer vision problems.


Research:

My research is concerned with learning representations for machine learning problems. My favorite machine learning models are hierarchical. My favorite problem domain is visual perception where I collaborate with the computer vision group of Thomas Brox. I still believe in the potential behind unsupervised (or weakly labeled) data and despise the black magic that is involved in training many neural network models; which I try to fight in collaboration with the research group of Frank Hutter. Apart from vision research I spend most of my days babysitting robots. Specifically, I try to invent learning algorithms that can learn representations enabling robots to perform complex tasks within the framework of reinforcement learning.

I am fascinated by the question of how our brain manages to learn powerful representations of the world that enable complex problem solving; and my long-term research goal is to make a small contribution to its answer.



Publications:

Short Papers and Pre-Prints:
  • Striving for Simplicity: The All Convolutional Net.
    J. T. Springenberg*, A. Dosovitskiy*, T. Brox and M. Riedmiller.
    arXiv:1412.6806, 2014
    (arXiv) (also appeared as a workshop paper at ICLR 2015)
  • Extrapolating Learning Curves of Deep Neural Networks.
    T. Domhan, J. T. Springenberg and F. Hutter.
    ICML AutoML Workshop, 2014
    (pdf) superseeded by the IJCAI 2015 paper
  • Unsupervised feature learning by augmenting single images.
    A. Dosovitskiy, J. T. Springenberg and T. Brox.
    arXiv:1312.5242 / ICLR Workshop Track, 2014
    (arXiv) superseeded by the NIPS 2014 paper

Full Papers:
  • *NEW* Asynchronous Stochastic Gradient MCMC with Elastic Coupling
    J. T. Springenberg, A. Klein, S. Falkner, F. Hutter
    arXiv:1612.00767, 2016/17
    (arXiv)
  • *NEW* Learning Curve Prediction with Bayesian Neural Networks
    A. Klein, J. T. Springenberg, S. Falkner, F. Hutter
    Currently under open review, 2017
    (link)
  • *NEW* Bayesian Optimization with Robust Bayesian Neural Networks
    J. T. Springenberg, A. Klein, S. Falkner, F. Hutter
    Neural Information Processing Systems (NIPS), 2016
    (link) (accepted as oral presentation)
  • *NEW* Deep Reinforcement Learning with Successor Features for Navigation across Similar Environments
    J, Zhang, J. T. Springenberg, J. Boedecker, W. Burgard
    (in submission)
  • Unsupervised and Semi-supervised Learning with Categorical Generative Adversarial Networks
    J. T. Springenberg
    International Conference on Learning Representations (ICLR), 2016
    (arXiv)
  • Embed to Control: A Locally Linear Latent Dynamics Model for Control from Raw Images
    M. Watter*, J. T. Springenberg*, J. Boedecker and M. Riedmiller.
    Neural Information Processing Systems (NIPS), 2015
    (pdf)
  • *NEW* Efficient and Robust Automated Machine Learning
    M. Feurer, A. Klein, K. Eggensperger, J. T. Springenberg, M. Blum and F. Hutter.
    Neural Information Processing Systems (NIPS), 2015
    (pdf)
  • Multimodal Deep Learning for Robust RGB-D Object Recognition.
    A. Eitel, J. T. Springenberg, L. Spinello, M. Riedmiller, W. Burgard.
    Proceedings of the IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), 2015
    (pdf)
  • Speeding up Automatic Hyperparameter Optimization of Deep Neural Networks by Extrapolation of Learning Curves.
    T. Domhan, J. T. Springenberg and F. Hutter.
    Proceedings of the 24th International Joint Conference on Artificial Intelligence (IJCAI), 2015
    (pdf)
  • Learning to Generate Chairs with Convolutional Neural Networks.
    A. Dosovitskiy, J. T. Springenberg and T. Brox.
    IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), 2015
    (url)
  • 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.
    W. Böhmer, J. T. Springenberg, J. Boedecker, M. Riedmiller, K. Obermayer.
    KI - Künstliche Intelligenz pp. 1-10. Springer, 2015
    (doi) (pdf)
  • Initializing Bayesian Hyperparameter Optimization via Meta-Learning.
    M. Feurer, J. T. Springenberg and F. Hutter.
    Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence (AAAI), 2015
    (pdf)
  • Discriminative Unsupervised Feature Learning with Convolutional Neural Networks.
    A. Dosovitskiy, J. T. Springenberg, M. Riedmiller and T. Brox.
    Neural Information Processing Systems (NIPS), 2014
    (link)
  • Approximate Real-Time Optimal Control Based on Sparse Gaussian Process Models.
    J. Boedecker, J. T. Springenberg, J. Wuelfing and M. Riedmiller.
    Adaptive Dynamic Programming and Reinforcement Learning (ADPRL), 2014
    (pdf)
  • Using Meta-Learning to Initialize Bayesian Optimization of Hyperparameters.
    M. Feurer, J. T. Springenberg and F. Hutter.
    ECAI workshop on Metalearning and Algorithm Selection (MetaSel), 2014
    (pdf) superseeded by the AAAI 2015 paper
  • Improving Deep Neural Networks with Probabilistic Maxout Units.
    J. T. Springenberg and M. Riedmiller.
    arXiv:1312.6116 / ICLR Workshop Track, 2014
    (arXiv)
  • Learning temporal coherent features through life-time sparsity.
    J. T. Springenberg and M. Riedmiller.
    International Conference on Neural Information Processing (ICONIP), 2012.
    (pdf)
  • A Learned Feature Descriptor for Object Recognition in RGB-D Data.
    M. Blum, J. T. Springenberg, J. Wülfing and M. Riedmiller.
    IEEE International Conference on Robotics and Automation (ICRA), 2012.
    (pdf)
  • On the Applicability of Unsupervised Feature Learning for Object Recognition in RGB-D Data.
    M. Blum, J. T. Springenberg, J. Wülfing and M. Riedmiller.
    NIPS Workshop on Deep Learning and Unsupervised Feature Learning, 2011.
    (pdf)
Theses:
  • Feature Learning using Temporal Coherence.
    J. T. Springenberg. Master Thesis.
    (pdf)
  • Machine Learning on Massively Parallel Architectures - A Case Study.
    J. T. Springenberg. Bachelor Thesis.
    University of Osnabrueck 09/2009.
    (pdf)

Disclaimer | Imprint