Research Fields

The chair of Big Data Analytics in Transportation has multiple research interests.

Note: These pages are currently under construction and are therefore not (yet) complete.

Algorithmization

In this research area, we look into the interaction between algorithms and their impact on society. As such, we are interested in answering questions to the following three topics:

• How are (artificially intelligent) algorithms used in social interaction and what are their effects?
• How do societies (including the general public and political and social elites) respond to the overt and covert influence of algorithms and algorithmic governance/management?
• How can the individual, social [and or societal], economic, and political impacts of algorithms be regulated?

This research is part of the Topical Program on Algorithmization and Social Interaction, which is headed by researchers from the University of Münster (Germany).

Automated Algorithm Selection

For many optimization problems, including machine learning, there are a great number of promising approaches. However, the performance of an approach may greatly depend on the particular problem instance.
Read more: Automated Algorithm Selection

Benchmarking

Benchmarking is an important procedure to better understand strengths and weaknesses of algorithms, studying the similarities between artificial and real-world problems, as well as investigating interactions between the problem instances and the algorithms executed thereon.
This line of research covers topics such as (statistical and/or visual) characterization of problem instances, generation of test problems, execution and comparison of numerous algorithms, performance assessment of algorithmic performances, etc.
Read more:  Benchmarking Network

Big Data Analytics

For almost all our research, we rely on data-driven approaches, which help us to extract valuable information from a variety of (heterogenous) data sources. Those data sources can vary in many ways: starting with classical tabular data (like the ones produced by MS Excel), followed by graphs and images, up to spatio-temporal data, as well as data streams.
In order to deal with the diverse types of data in an appropriate way, we make use of a variety of suitable methods and algorithms. These range from highly efficient methods for data (pre-)processing, as well as statistical and/or visual methods for informative data exploration, to the usage of statistical and/or machine learning methods (incl. deep learning neural networks) for identifying patterns from the data.

Evolutionary Computation

For practical applications, deterministic optimization algorithms can be very inefficient (e.g., because the underlying assumptions can't be met, or the problem at hand is a black-box). In such cases, the usage of randomized search heuristics (such as CMA-ES) can be beneficial.
Here, we are interested in issues such as

• a better understanding of the strengths and weaknesses of various algorithms,
• finding out how they can be effectively hybridized (within a powerful metaheuristic), or
• how certain components of the search heuristic could be efficiently adapted to the problem at hand.

Exploratory Landscape Analysis

Read more: Exploratory Landscape Analysis

Large-Scale Machine Learning and/or Optimization

As large-scale problems are hardly visualizable, algorithms are usually designed based on low-dimensional problems. Subsequently, they are executed on large-scale problems, in the hope that their (search and/or decision) behavior will transfer correspondingly.
Here, we are interested in examining whether this assumption is actually true, or how large-scale problems differ from low-dimensional ones. Further, we study how large-scale problems can efficiently be reduced to low-dimensional problems without losing relevant information.

Machine Learning

In the context of machine learning (ML), we are interested in a variety of topics that either help to better understand the mechanics of the underlying algorithms, or which help to facilitate their usage within practical applications.
As such, we are particulary interested in the following subareas of ML:

• automated machine learning (AutoML)
• interpretable machine learning (IML)
• adversarial attacks, and especially how to make algorithms more robust against them
• uncertainty of algorithmic decisions (and hence the trustworthiness into their models)
• combination of machine learning and optimization:
  • automated algorithm selection
    → using ML for more efficient optimization, or
  • (hyper-)parameter optimization
    → using optimization for better ML models

Multi-Objective Optimization

Read more: Multi-Objective Optimization

Route Optimization:

Read more: Route Optimization