Learning capability represents a central component of intelligence. Machine learning is the subfield of artificial intelligence that deals with algorithms that are capable of learning. These algorithms independently learn to recognize patterns and regularities in data and use them to make predictions and decisions. Deep learning is a prominent subfield of machine learning that uses artificial neural networks to model patterns and regularities, and which has led to breakthroughs in numerous application areas such as font, image, and speech recognition over the past decade. Recent artificial intelligence milestones, such as the victory of the Go program AlphaGo in a match against the world’s best human player, can also be attributed to artificial neural networks.
Within deep learning, the team of Prof. Dr. Asja Fischer mainly conducts research in the area of analysis and development of probabilistic methods and models. Thereby, a focus is on the theory underpinning the methods. Despite the immense success of deep learning algorithms in many application areas, relatively little is known about their theoretical properties. However, a better understanding of the mathematical properties is of enormous importance to make these algorithms more robust, reliable, interpretable, and secure. However, the group also conducts research on the application of neural networks to exciting practical problems and in interdisciplinary projects.
The following areas of focus have been identified in the research to date:
- development and analysis of generative modelsand MCMC-based learning methods
- analysis of the optimization methods of neural networks
- uncertainty estimation methods for neural network predictions
- biological plausible deep learning
- knowledge graph analysis and deep learning methods for question answering over knowledge graphs
- machine learning for IT security and secure machine learning