Jörg Ontrup

Large-scale data exploration with the hierarchically growing hyperbolic SOM

We introduce the Hierarchically Growing Hyperbolic Self-Organizing Map (H2SOM) featuring two extensions of the HSOM (hyperbolic SOM): (i) a hierarchically growing variant that allows for incremental training with an automated adaptation of lattice size to achieve a prescribed quantization error and (ii) an approximate best match search that utilizes the special structure of the hyperbolic lattice to achieve a tremendous speed-up for large map sizes. Using the MNIST and the Reuters-21578 database as benchmark datasets, we show that the H2SOM yields a highly efficient visualization algorithm that combines the virtues of the SOM with extremely rapid training and low quantization and classification errors.

Semi-automated image analysis for the assessment of megafaunal densities at the Arctic deep-sea observatory HAUSGARTEN

Megafauna play an important role in benthic ecosystem function and are sensitive indicators of environmental change. Non-invasive monitoring of benthic communities can be accomplished by seafloor imaging. However, manual quantification of megafauna in images is labor-intensive and therefore, this organism size class is often neglected in ecosystem studies. Automated image analysis has been proposed as a possible approach to such analysis, but the heterogeneity of megafaunal communities poses a non-trivial challenge for such automated techniques. Here, the potential of a generalized object detection architecture, referred to as iSIS (intelligent Screening of underwater Image Sequences), for the quantification of a heterogenous group of megafauna taxa is investigated. The iSIS system is tuned for a particular image sequence (i.e. a transect) using a small subset of the images, in which megafauna taxa positions were previously marked by an expert. To investigate the potential of iSIS and compare its results with those obtained from human experts, a group of eight different taxa from one camera transect of seafloor images taken at the Arctic deep-sea observatory HAUSGARTEN is used. The results show that inter- and intra-observer agreements of human experts exhibit considerable variation between the species, with a similar degree of variation apparent in the automatically derived results obtained by iSIS. Whilst some taxa (e. g. Bathycrinus stalks, Kolga hyalina, small white sea anemone) were well detected by iSIS (i. e. overall Sensitivity: 87%, overall Positive Predictive Value: 67%), some taxa such as the small sea cucumber Elpidia heckeri remain challenging, for both human observers and iSIS.

Biigle - Web 2.0 enabled labelling and exploring of images from the Arctic deep-sea observatory HAUSGARTEN

Deep-sea research relies strongly on the use of high-resolution cameras which generate large quantities of footage. The material can currently, however, often not be used to its full potential as the analysis is time- and labour-intensive and requires the input of many different taxonomic experts. Here, we present a system which enables the collaboration of experts from various places and the application of machine-vision tools on footage from the Arctic deep-sea observatory, HAUSGARTEN. Biigle (Bielefeld Image Graphical Labeller and Explorer) is a Web 2.0 based platform containing easily uploaded images that can be accessed by collaborating scientists. Since Biigle is realised as a rich internet application, there is no need for the local installation of complex software packages. The scientists can use a standard web browser to access the image database and immediately explore or label images. They have instant access to the data submitted by other scientists and are directly involved in the emerging community. Biigle also offers an application interface for machine-vision components aiming at the automated analysis of seafloor images. As a first module, a laser point detection allows for an automated calibration of the area covered by the camera which is vital to derive faunal density estimates. The laser points were detected in all but eight of 1883 images tested in total. The combination of human expert labels and machine-vision results can be exported into spreadsheets offering a well-established standard for further data analyses. Biigle can be accessed at http://www.biigle.de with the username and the password “test” for testing purposes.

A computational feature binding model of human texture perception

We present a computational model for human texture perception which assigns functional principles to the Gestalt laws of similarity and proximity. Motivated by early vision mechanisms, in the first stage, local texture features are extracted by utilizing multi-scale filtering and nonlinear spatial pooling. In the second stage, features are grouped according to the spatial feature binding model of the competitive layer model (CLM; Wersing et al. 2001). The CLM uses cooperative and competitive interactions in a recurrent network, where binding is expressed by the layer-wise coactivation of feature-representing neurons. The Gestalt law of similarity is expressed by a non-Euclidean distance measure in the abstract feature space with proximity being taken into account by a spatial component. To choose the stimulus dimensions which allow the most salient similarity-based texture segmentation, the feature similarity metrics is reduced to the directions of maximum variance. We show that our combined texture feature extraction and binding model performs segmentation in strong conformity with human perception. The examples range from classical microtextures and Brodatz textures to other classical Gestalt stimuli, which offer a new perspective on the role of texture for more abstract similarity grouping.

Hyperbolic SOM-based clustering of DNA fragment features for taxonomic visualization and classification

MOTIVATION Modern high-throughput sequencing technologies enable the simultaneous analysis of organisms in an environment. The analysis of species diversity and the binning of DNA fragments of non-sequenced species for assembly are two major challenges in sequence analysis. To achieve reasonable binnings and classifications, DNA fragment structure has to be represented appropriately, so it can be processed by machine learning algorithms. RESULTS Hierarchically growing hyperbolic Self-Organizing maps (H(2)SOMs) are trained to cluster small variable-length DNA fragments (0.2-50 kb) of 350 prokaryotic organisms at six taxonomic ranks Superkingdom, Phylum, Class, Order, Genus and Species in the Tree of Life. DNA fragments are mapped to three different types of feature vectors based on the genomic signature: basic features, features considering the importance of oligonucleotide patterns as well as contrast enhanced features. The H (2)SOM classifier achieves high classification rates while at the same time its visualization allows further insights into the projected data and has the potential to support binning of short sequence reads, because DNA fragments can be grouped into phylogenetic groups. AVAILABILITY An implementation of the H(2)HSOM classifier in Matlab is provided at www.techfak.uni-bielefeld.de/ags/ani/projects/HHSOMSeqData.

Interactive visualization and navigation in large data collections using the hyperbolic space

We propose the combination of two recently introduced methods for the interactive visual data mining of large collections of data. Both hyperbolic multidimensional scaling (HMDS) and hyperbolic self-organizing maps (HSOM) employ the extraordinary advantages of the hyperbolic plane (H2): (i) the underlying space grows exponentially with its radius around each point deal for embedding high-dimensional (or hierarchical) data; (ii) the Poincare model of the IH/sup 2/ exhibits a fish-eye perspective with a focus area and a context preserving surrounding; (in) the mouse binding of focus-transfer allows intuitive interactive navigation. The HMDS approach extends multidimensional scaling and generates a spatial embedding of the data representing their dissimilarity structure as faithfully as possible. It is very suitable for interactive browsing of data object collections, but calls for batch precomputation for larger collection sizes. The HSOM is an extension of Kohonens self-organizing map and generates a partitioning of the data collection assigned to an IH/sup 2/ tessellating grid. While the algorithms complexity is linear in the collection size, the data browsing is rigidly bound to the underlying grid. By integrating the two approaches, we gain the synergetic effect of adding advantages of both. And the hybrid architecture uses consistently the IH/sup 2/ visualization and navigation concept. We present the successfully application to a text mining example involving the Reuters-21578 text corpus.

Megafaunal assemblages from two shelf stations west of Svalbard

Abstract Megafauna plays an important role in benthic ecosystems and contributes significantly to benthic biomass in the Arctic. The distribution is mostly studied using towed cameras. Here, we compare the megafauna from two sites located at different distances from the Kongsfjord: one station at the entrance to the fjord, another on the outer shelf. Although they are only located 25 km apart and at comparable depth, there were significant differences in their species composition. While the inshore station was characterized by shrimps (2.57±2.18 ind. m−2) and brittlestars (3.21± 3.21 ind. m−2), the offshore site harboured even higher brittlestar densities (15.23±9.32 ind. m−2) and high numbers of the sea urchin Strongylocentrotus pallidus (1.23±1.09 ind. m−2). Phytodetrital concentrations of the upper sediment centimetres were significantly higher inshore compared with offshore. At a smaller scale, there were also differences in the composition of different transect sections. Several taxa were characterized by a patchy distribution along transects. We conclude that these differences were caused primarily by habitat characteristics. The seafloor inshore was characterized by glacial soft sediments, whereas the station offshore harboured large quantities of stones. Although the use of a new web-2.0-based tool, BIIGLE (http://www.BIIGLE.de), allowed us to analyse more images (∼90) than could have been achieved by hand, taxon area curves indicated that the number of images analysed was not sufficient to capture the species inventory fully. New automated image analysis tools would enable a rapid analysis of larger quantities of camera footage.

Detecting, Assessing and Monitoring Relevant Topics in Virtual Information Environments

The ability to assess the relevance of topics and related sources in information-rich environments is a key to success when scanning business environments. This paper introduces a hybrid system to support managerial information gathering. The system is made up of three components: 1) a hierarchical hyperbolic SOM for structuring the information environment and visualizing the intensity of news activity with respect to identified topics, 2) a spreading activation network for the selection of the most relevant information sources with respect to an already existing knowledge infrastructure, and 3) measures of interestingness for association rules as well as statistical testing facilitates the monitoring of already identified topics. Embedding the system by a framework describing three modes of human information seeking behavior endorses an active organization, exploration and selection of information that matches the needs of decision makers in all stages of the information gathering process. By applying our system in the domain of the hotel industry we demonstrate how typical information gathering tasks are supported. Moreover, we present an empirical study investigating the effectiveness and efficiency of the visualization framework of our system.

The Cyborg Astrobiologist: first field experience

29 pages, 10 figures.-- Final editor version available at: http://dx.doi.org/10.1017/S147355040500220X

The Cyborg Astrobiologist: testing a novelty detection algorithm on two mobile exploration systems at Rivas Vaciamadrid in Spain and at the Mars Desert Research Station in Utah

In previous work, a platform was developed for testing computer-vision algorithms for robotic planetary exploration. This platform consisted of a digital video camera connected to a wearable computer for real-time processing of images at geological and astrobiological field sites. The real-time processing included image segmentation and the generation of interest points based upon uncommonness in the segmentation maps. Also in previous work, this platform for testing computer-vision algorithms has been ported to a more ergonomic alternative platform, consisting of a phone camera connected via the Global System for Mobile Communications (GSM) network to a remote-server computer. The wearable-computer platform has been tested at geological and astrobiological field sites in Spain (Rivas Vaciamadrid and Riba de Santiuste), and the phone camera has been tested at a geological field site in Malta. In this work, we (i) apply a Hopfield neural-network algorithm for novelty detection based upon colour, (ii) integrate a field-capable digital microscope on the wearable computer platform, (iii) test this novelty detection with the digital microscope at Rivas Vaciamadrid, (iv) develop a Bluetooth communication mode for the phone-camera platform, in order to allow access to a mobile processing Computer at the field sites, and (v) test the novelty detection on the Bluetooth-enabled phone camera connected to a netbook Computer at the Mars Desert Research Station in Utah. This systems engineering and field testing have together allowed Lis to develop a real-time computer-vision system that is capable, for example, of identifying lichens as novel within a series of images acquired in semi-arid desert environments. We acquired sequences of images of geologic outcrops in Utah and Spain consisting of various rock types and colours to test this algorithm. The algorithm robustly recognized previously observed units by their colour, while requiring only a single image or a few images to learn colours as familiar, demonstrating its fast learning capability.