The Statistics and Data Analysis Section has special emphasis on statistics, quantitative genomics, bioinformatics, pattern recognition and software development. The section is dedicated to supporting other departments at DTU and its external partners with skills, knowledge, and consultancy within the field of statistics and data analysis - including biological data.
DTU Compute has a long history of conducting statistical consultancy and participating in research and publication collaborations with other departments at DTU, other universities and external partners. The basic purpose of the section is to formalize this collaboration and contribute to even better research and public service sector consultancy at DTU. With its close connections to DTU Compute research groups in cognitive systems, image analysis and scientific computing, the section stands as a dynamic and substantial player both locally in the Danish statistics arena and internationally.
The section aims at strengthening the connection to the statistics, bioinformatics, and computational biology related activities in other departments at DTU and championing the use of high-quality statistics within the Public Sector Consultancy activities at DTU. The department already hosts internal and external university consultancy services in statistics.
Statistical process control (SPC) is the general methodology based on statistical methods that can be used in monitoring and control of product and/or process quality. We often expect the processes and products to exhibit variability. This inherent variability in a process is defined as the variability that occurs by chance. If a process exhibits such variation, it is said to be in a state of statistical control. However, when there is an unusual and/or unexpected occurrence of excess variability due to an assignable cause, the process is said to be out of statistical control. This could for example be due to wear and tear, defective/impure raw materials, errors in data collection schemes, etc. Statistical process control aims to identify these assignable causes generating the excessive variation and help to bring the process back in statistical control.
Murat Külahci Professor
We deal with the planning of experiments where variation is present. Often someone planning to do an experiment is interested in the effect of some process or intervention on some experimental unit. Given experimental conditions, the main challenge is to formulate experimental plans which will provide informative data suitable for statistical analysis. Design of experiments is a discipline that has very broad application across all the natural and social sciences. Our research areas include design of computer experiments, design of animal experiments and the design of split plot experiments.
Murat Külahci Professor
The use of humans as measurement instruments is playing an increasing role in product development and user driven innovation in many engineering industries. The global food industry frequently uses sensory and consumer data as the basis for decision making. This calls for improved understanding of both the instrument itself and the modelling and empirical treatment of data. This approach ranges from experimental psychology to mathematical modelling and statistics combined with specific product knowledge of everything from food, TVs, Hearing aids to mobile phones etc. The field of chemometrics uses mathematical, statistical and other methods to design or select optimal measurement procedures and experiments to provide maximum relevant chemical information by analyzing chemical data.
Quantitative genomics is closely related to population genetics, statistical genetics and genetic epidemiology. This area of research is concerned with diseases and traits that have a complex or mixed inheritance background, influenced by external environmental effects and their interactions with genes. In this research, we estimate population genetic parameters such as heritability, genetic correlation and variance components for quantitative traits and diseases in animals and humans using population-level genetics/genomics and epidemiological and environmental data.
Modern statistical models are used in statistical learning and statistical engineering to analyse the increased amounts of data collected everywhere in our society on a daily basis. The methods include random forests, regularisation strategies, sparse methods, support vector machines, boot strapping, deep belief networks, and any more. The field is very active with a huge number of journal papers published. The research can be applied through automation strategies in industrial productions, in educational measurements, in bioinformatics, gene studies and in management an business intelligence.
Line Katrine Harder Clemmensen Professor
Just a few years ago learning was all about going to a lecture and listen to the teacher. Today learning is very much more and takes place on many different learning platforms. Especially e-learning is in focus and provides lots of possibilities for making learning deeper and more individual, innovative and fun. It is crucial that learning methods are developed in cooperation with the area it is used in. Our focus is on learning which enhances the expertise of the entire department to develop the learning technologies for tomorrow’s engineering education. We have recently developed and implemented a new approach to e-learning called HEROS: a non-linear learning object based learning system. It lets students design their own individual research based course in statistics and mathematics in a way which fits their individual levels and learning styles.
The research in this field is focused on systems with discrete state space like queueing and inventory systems. More technically we specialize in the field of matrix analytic methods which was popularized by Marcel Neuts in the 1970'es and onwards. Current applications have been within call center modeling and modeling of marine systems. Most recently we are interested in parameter estimation and statistical testing in multivariate matrix-exponential distributions. These methods can be applied in design of communications systems, to determine software reliability, in call center modelling and design and in statistical modelling etc.
Bo Friis Nielsen Professor
