The Pattern Recognition & Bioinformatics Section is a organizational section focussed on pattern recognition and its applications to computer vision and bioinformatics.
Pattern recognition is concerned with processing raw measurement data by a computer to arrive at a prediction, which can then be used to formulate a decision or action to take. Problems to which pattern recognition are applied have in common that they are too complex to model explicitly, thus requiring algorithms to learn parameters in generic models from limited sets of examples. Pattern recognition practice is firmly focused on real-world, sensor-based applications. This places it at the core of the current process of scientific discovery, by allowing researchers to derive regularities in large amounts of data in areas as diverse as physics, biology and geology, but also psychology and neuroscience. Pattern recognition algorithms also find application in industrial and consumer settings, allowing machines to sense the environment and to decide on actions or support human decision making. The PRB section studies both aspects in three different research labs. One research lab (pattern recognition lab) focuses on the foundations of pattern recognition: representation and generalization, in which new ways of describing objects and learning from examples are studied. Two other research labs (the computer vision lab and the bioinformatics lab) apply these techniques in the domains of images and of molecular biology.
Research activities & vision
The Pattern Recognition Lab (http://prlab.tudelft.nl/) is concerned with the classical trinity of representation, generalization, and evaluation, the core elements of every pattern recognition system. The principal focus is on developing tools and theories and gaining knowledge and understanding applicable to a broad range of general problems. Typically this involves sensory data, e.g. time signals, images, video streams, or other physical measurement data. The lab has significantly contributed to the field, mainly on neural networks, the use of dissimilarities and one-class classifiers. Current major research directions are dissimilarity-based pattern recognition, multiple classifier systems, and multiple instance learning. In future, research will focus on alternative evaluation functions, e.g. ROC analyses, and pattern recognition in particular settings, such as semi-supervised and active learning.
The Delft Bioinformatics Lab (http://bioinformatics.tudelft.nl/) deals with developing novel computer models and algorithms to further fundamental biological knowledge and apply these models and algorithms to advance the state-of-the-art in health care and industry. The lab focuses on data-driven bioinformatics: creating algorithms to infer and exploit simple models of complex interactions, by coupling biological insights and available prior knowledge to high-throughput measurements. Recent contributions include the discovery of novel cancer genes by analyzing and modeling insertional mutagenesis data; proposing (combinatorial) cultivation dependent transcription factor activities based on a decomposition of transcriptomics data; and assessing gene therapy protocols by integrating viral insertions with gene expression data. Future prominent areas include studying robustness in microbial systems, application of scale-space theory to detect functional modules in molecular data at different levels, analyzing same-sample multi-modal molecular data to infer combinatorial interactions between different components, and building atlases for both local sequence variations and gross structural variants during evolution experiments.
The Computer Vision Lab (http://visionlab.tudelft.nl/) focuses on the segmentation and analysis of multidimensional sensor data (image sequences, multiple cameras, 3D/4D medical data like MRI and CT). Segmentation is used in the broadest possible sense, i.e. distinguishing any relevant image information from non-relevant information. The main research areas currently covered are 3D imaging (camera calibration, disparity estimation, 3D reconstruction, 3DTV/free viewpoint rendering), biomedical imaging (medical image segmentation, 2D/3D model reconstruction), social or human signal processing (pose and gesture recognition and tracking) and surveillance (video object detection, recognition and tracking). In the future, biomedical imaging will gain importance (by intensifying the collaboration with the Image Processing Division of the Leiden University Medical Center) and will be coupled with molecular data.