At the Institute for Information Processing (TNT), methods in the broad field of machine learning are developed and applied to various problems. The goal is to automatically extract knowledge and semantic relationships from large datasets. This information is extremely important for applications such as autonomous driving, cancer diagnosis, aerial image analysis, augmented reality, and Industry 4.0.
Autonomous vehicles increase the safety of all participants in road traffic by replacing fallible human drivers with reliable algorithms. Robust methods for semantic analysis of the driving environment are necessary to detect hazardous situations and make decisions based on them. Using machine learning techniques, TNT develops algorithms that analyze data from various sensors, such as cameras or Lidar, to enable safe autonomous driving.
Using adversarial networks, it is possible to accomplish various tasks such as generating realistic-looking images and transferring information between different domains and sensors. For example, the resolution of an image can be increased, or missing or obscured image areas can be reconstructed.
Participants in everyday road traffic can use low-cost sensors to quickly collect and make large amounts of interesting data accessible. From this data, semantic information is extracted using graphical models and specialized neural networks, such as construction sites that restrict movement or peak traffic times.
Video games provide diverse but at the same time well-controlled environments for researching decision-making algorithms. For instance, using reinforcement learning, intelligent agents for video games are developed that have the same information and opportunities as real players.
A lot of interesting information can be extracted from aerial images, such as current maps, city growth over time, the assessment of traffic volumes, or the occupancy of parking spaces. TNT uses deep learning to automatically determine land use types and recognize objects in aerial images.
Neural Networks (CNNs, Autoencoders, RNNs, GANs, …), Statistical Learning Methods (Random Forest, Gaussian Mixture Models, Hidden Markov Models, …), Reinforcement Learning (Q-Learning, MCTS, …)
