Supervisor: Prof. Ville Kyrki (firstname.lastname@example.org)
Advisor: MSc. Oliver Struckmeier (email@example.com)
Keywords: Representation Learning, Neurorobotics, Slow Features, Robot Navigation
Representing data is crucial in many information processing tasks. Usually a “good” representation of data makes the learning of a down-stream task easier. The beta-Variational Autoencoder is an established method of extracting latent representations . In our current work we investigate a new method to extract slowly varying features using a VAE from time correlated signals such as videos and to use them to more efficiently train down-stream tasks.
Recent research in neuroscience  has suggested that slow features and sparse coding can lead to the self-formation of complex cell structures like place-, head-direction- and spatial view cells. In recent research we presented ViTa-SLAM  a brain-inspired navigation framework based on grid and place cells in the rodent brain.
The goal of this thesis to bridge the gap between slow representation learning and brain-inspired navigation. After reviewing the relevant literature from both domains, an experiment will investigate if slow representations can lead to the self organization of structures similar to complex cells in the brain. If the results confirm the hypothesis, another experiment will investigate if the emerging structures can be used for navigation similar to the hand-crafted place-cell network in ViTa-SLAM.
- Review state of the art literature
- Design an experiment to investigate the self-organization of grid and place cells from slow representations
- Design an experiment in which the emerging structures are used for navigation
- Evaluation of the experiment results
- Pre-requisites: Good Python skills, Deep Learning
- Tools: PyTorch, Robot Operating System (ROS) for the navigation experiment
- Start: Available immediately
- Possibility to expand on the Master thesis and publish a paper
-  Struckmeier, Oliver, et al. “ViTa-SLAM: A Bio-inspired Visuo-Tactile SLAM for Navigation while Interacting with Aliased Environments.” 2019 IEEE International Conference on Cyborg and Bionic Systems (CBS). IEEE, 2019.
-  Franzius, M., Sprekeler, H., Wiskott, L., 2007. Slowness and Sparseness Lead to Place, Head-Direction, and Spatial-View Cells. PLoS Comput Biol 3, e166. https://doi.org/10.1371/journal.pcbi.0030166
-  Higgins, I., Matthey, L., Pal, A., Burgess, C., Glorot, X., Botvinick, M., Mohamed, S., Lerchner, A., 2017. β-VAE: LEARNING BASIC VISUAL CONCEPTS WITH A CONSTRAINED VARIATIONAL FRAMEWORK 13.