Research areas and PhD topics

The most important research emerges form the interaction of fundamental theory with applications. We develop theoretically sound methods that solve important problems in emerging areas. Here is a list of topics we study, and the areas where we apply them.

Fundamental techniques and theory

Machine learning, data mining and algorithms

For machine learning in critical applications, it is vitally important that the learning does not consist of arbitrary errors. In such cases, we need guarantees on the quality and confidence of the learning.

Our research therefore focuses on provably good and fast techniques for learning, with a particular focus on complex data such as networks and trajectories. The following are some interesting challeneges in this area

Privacy preserving machine learning: statistical privacy

In the age of big data, the privacy of individual information is increasingly important. Information does not only leak through hacking and data sharing, critical information can also be inferred from the output behavior of a machine learning application, without knowledge of the input.

Preventing such inference about individuals in the dataset is the idea behind statistical privacy techniques like differential privacy. It is an interesting topic that involves deep understanding of computation and machine learning to ensure precisely that statistical patterns can be learned while individual information will be hidden. This topic still has a long way to go before privacy becomes ubiquitous and effective. It is critical not just for the development of machine learning, but also in general.

Computational geometry and topology

Geometry and topology are playing increasingly important roles in relation to data science and machine learning, as we get more data with locations, embeddings and shapes. We are interested in the general question of the "shape of data" interpreted using topology, hyperbolic geometry, and various other structures.

Application areas and systems

Sensor networks and IoT

In the distributed environment of sensor and IoT networks, computation takes on a unique character. We work on a mixture of geometry, ML and privacy and develop fast information algorithms.

Self driving cars and robotics

These are some of the hottest emerging areas. They need a variety of technologies to operate — ranging from mechanical engineering to maths. We use machine learning, geometry, topology for motion planning, prediction, motion analysis and various other problems.

Mobility analytics and management, urban data science, smart cities

With publicly available GPS datasets, we can now dive into the behavior and patterns of mobility in cities. Making sense of this data is a fundamental problem in smart cities and related topics. It is however, also very difficult question. What constitutes a “pattern”? What can we really hope to learn? Where can we improve things? We use ML, geometry, topology and various data mining techniques to investigate these issues. And of course, privacy is a critical issue in this topic.

Social networks and network analysis

Analysis of networks has a far reach, touching on many areas of computer science and beyond. We focus on social network, road networks and various other topics, and use algorithms, geometry and graph embedding techniques.