On Integrating Data Mining into Business Processes.

SVM Classifier Estimation from Group Probabilities.
A learning problem that has only recently gained attention in the machine learning community is that of learning a classifier from group probabilities. It is a learning task that lies somewhere between the well-known tasks of supervised and unsupervised learning, in the sense that for a set of observations we do not know the labels, but for some groups of observations, the frequency distribution of the label is known. This learning problem has important practical applications, for example in privacy-preserving data mining. This paper presents an approach to learn a classifier from group probabilities based on support vector regression and the idea of inverting a classifier calibration process. A detailed analysis will show that this new approach outperforms existing approaches. 1.
Privacy%2DPreserving Data%2DMining.
In this paper we address the issue of privacy preserving data mining. Specifically, we consider a  scenario in which two parties owning confidential databases wish to run a data mining algorithm on  the union of their databases, without revealing any unnecessary information. Our work is motivated  by the need to both protect privileged information and enable its use for research or other purposes. The
Building a System for Advancing Clinico%2DGenomic Trials on Cancer.
Abstract The analysis of clinico-genomic data poses complex computational problems. In the project ACGT, a grid-based software system to support clinicians and bio-statisticians in their daily work is being developed. Starting with a detailed user requirements analysis, and with the continuous integration of usability analysis in the development process, the project strives to develop an architecture that will substantially improve the way clinico-genomic trials are conducted today. In this paper, results of the initial requirements analysis and approaches to address these requirements are presented. We also discuss the importance of appropriate metadata to tailor the system to the needs of the users. 1
On Subgroup Discovery in Numerical Domains.

On subgroup discovery in numerical domains.

GridR An R%2Dbased tool for scientific data analysis in grid environments.
journal homepage: www.elsevier.com/locate/fgcs
Tight Optimistic Estimates for Fast Subgroup Discovery.
Abstract. Subgroup discovery is the task of finding subgroups of a population which exhibit both distributional unusualness and high generality. Due to the non monotonicity of the corresponding evaluation functions, standard pruning techniques cannot be used for subgroup discovery, requiring the use of optimistic estimate techniques instead. So far, however, optimistic estimate pruning has only been considered for the extremely simple case of a binary target attribute and up to now no attempt was made to move beyond suboptimal heuristic optimistic estimates. In this paper, we show that optimistic estimate pruning can be developed into a sound and highly effective pruning approach for subgroup discovery. Based on a precise definition of optimality we show that previous estimates have been tight only in special cases. Thereafter, we present tight optimistic estimates for the most popular binary and multi-class quality functions, and present a family of increasingly efficient approximations to these optimal functions. As we show in empirical experiments, the use of our newly proposed optimistic estimates can lead to a speed up of an order of magnitude compared to previous approaches. 1
A Semantic Grid Services Architecture in Support of Efficient Knowledge Discovery from Multilevel Clinical and Genomic Datasets.
This paper presents the architectural considerations of the Advancing Clinico-Genomic Trials on Cancer (ACGT) project aiming at delivering a European Biomedical Grid in support of efficient knowledge discovery in the context of post-genomic clinical trials on cancer. Our main research challenge in ACGT is the requirement to develop an infrastructure able to produce, use, and deploy knowledge as a basic element of advanced applications, which will mainly constitute a Biomedical Knowledge Grid. Our approach to offer semantic modelling of available services and data sources to support high level services and dynamic services for discovery and composition will be presented. In particular, ontologies and metadata are the basic elements through which Grid intelligence services can be developed, and the current achievmnets of the project in this domain will be discussed. 1
Supporting Parallel R Code in Clinical Trials A Grid%2DBased Approach.
In this paper, we describe an extension to the ACGT GridR environment which allows the parallelization of loops in R scripts in view of their distributed execution on a computational grid. The ACGT GridR service is extended by a component that uses a set of preprocessor-like directives to organize and distribute calculations. The use of parallelization directives as special R comments provides users with the potential to accelerate lengthy calculations with changes to preexisting code. The GridR service and its extension are developed as components of the ACGT platform, one aim of which is to facilitate the data mining of clinical trials involving large datasets. In ACGT, GridR scripts are executed in the framework of a specifically developed workflow environment, which is also briefly outlined in the present article. 1.
Benchmarking Domain Specific Processors A Case Study of Evaluating a Smart Card Processor Design.

Procurement Fraud Discovery using Similarity Measure Learning.
Abstract. This paper describes an approach to detect hints on procurement fraud. It was developed within the context of a European Union project on fraud prevention. Procurement fraud is a special kind of fraud that occurs when employees cheat their own employers by executing or triggering bogus payments. The approach presented here is based on the idea to learn a similarity measure that compares an employee (or payroll) standing-data record to a creditor record, in order to detect creditors that are suspiciously similar to employees. To this ends, it combines several simple similarity measures like address similarity or spacial similarity using a weighting scheme. The weights, that is the overall similarity function, are learned from user input specifying whether a particular pair of payroll and creditor data records are similar. Key words: similarity measure learning, fraud detection, health care 1
Facilitating Clinico%2DGenomic Knowledge Discovery by Automatic Selection of KDD Processes.
The analysis of clinico-genomic data poses complex problems for machine learning. As high volumes of data can be generated easily, selecting the most suitable KDD-process for the problem at hand becomes increasingly hard, even for experienced researchers. The main idea of this paper is to facilitate process selection by representing each data set by a graph based on the ontology that describes data set attributes, and to apply graph mining methods to perform a similarity search. Some new measures for an effective comparison of a data set graph induced from the ontology are proposed. The effectiveness of the proposed approach is evaluated on three datasets. The results show that using ontology-based characteristics leads to improving the characterization of a data set. 1.

Knowledge Discovery Scientific Workflows in Clinico%2DGenomics.


Robust Probabilistic Calibration.
Probabilistic calibration is the task of producing reliable estimates  of the conditional class probability P (class|observation) from  the outputs of numerical classifiers. A recent comparative study [1] revealed  that Isotonic Regression [2] and Platt Calibration [3] are most  e#ective probabilistic calibration technique for a wide range of classifiers.
Grid%2DBased Knowledge Discovery in Clinico%2DGenomic Data.
Knowledge discovery in clinico-genomic data is a task that  requires to integrate not only highly heterogeneous kinds of data, but also  the requirements and interests of very di#erent user groups. Technologies  of grid computing promise to be an e#ective tool to combine all these  requirements into a single architecture. In this paper, we describe scenarios  and future research directions related to grid-based knowledge discovery  in clinico-genomic data, and introduce the approach taken by the  recently launched ACGT project    . The whole endeavor is considered in  the context of biomedical informatics research and aims towards the realization  of an integrated and grid-enabled biomedical infrastructure. The  presented integrated clinico-genomics knowledge discovery (ICGKD) scenario  and its process realization is based on a multi-strategy data-mining  approach that seamlessly integrates three distinct data-mining components:  clustering, association rules mining, and feature-selection. Preliminary  experimental results are indicative of the rational and reliability of  the approach.
A Simple Method For Estimating Conditional Probabilities For SVMs.
Support Vector Machines (SVMs) have become  a popular learning algorithm, in particular for  large, high-dimensional classification problems.
Learning with Local Models.
In this contribution, we deal with active learning, which gives the learner the power to select training samples. We propose a novel query algorithm for local learning models, a class of learners that has not been considered in the context of active learning until now. Our query algorithm is based on the idea of selecting a query on the borderline of the actual classification. This is done by drawing on the geometrical properties of local models that typically induce a Voronoi tessellation on the input space, so that the Voronoi vertices of this tessellation offer themselves as prospective query points. The performance of the new query algorithm is tested on the two-spirals problem with promising results. Keywords: active learning, local models, query based learning, vector quantization  1 Introduction  In supervised learning, we are interested in training a student on a set of input-- output pairs generated by an unknown target function in such a way that the student does not only re...
Concept Drift and the Importance of Example.
For many learning tasks where data is collected over an extended period of time, its underlying distribution is likely to change. A typical example is information  ltering, i.e. the adaptive classi  cation of documents with respect to a particular user interest. Both the interest of the user and the document content change over time. A  ltering system should be able to adapt to such concept changes.
A Multistrategy Approach to the Classification of Phases in Business Cycles.
The classification of business cycles is currently performed using either macro-economic equations or linear discriminant analysis. It is a hard and important problem to classify in which economic phase we are in. Government as well as business decisions rely on the assessment of the current business cycle. In this paper, we investigate how economists can be better supported by a combination of machine learning techniques. We have successfully applied Inductive Logic Programming (ILP). The application of ILP requires pre-processing in order to establish time and value intervals. To this end, top-down induction of decision trees is used. The rule sets learned from different experiments were analysed with respect to correlations in order to find a concept drift or shift.
Support Vector Machines in Relational Databases.
Today, most of the data in business applications is stored in relational databases.
Support Vector Machines in Relational Databases.
Today, most of the data in business applications is stored in relational databases.
Incremental Learning with Support Vector Machines.
Support Vector Machines (SVMs) have become a popular tool for learning with large amounts of high dimensional data. However, it may sometimes be preferable to learn incrementally from previous SVM results, as computing a SVM is very costly in terms of time and memory consumption or because the SVM may be used in an online learning setting. In this paper an approach for incremental learning with Support Vector Machines is presented, that improves existing approaches. Empirical evidence is given to prove that this approach can effectively deal with changes in the target concept that are results of the incremental learning setting.
SOM accelerator system.

A High Performance SOFM Hardware%2DSystem.
Many applications of Selforganizing Feature Maps (SOFMs) need a high performance hardware system in order to be efficient. Because of the regular and modular structure of SOFMs, a hardware realization is obvious. Based on the idea of a massively parallel system, several chips have been designed, manufactured and tested by the authors. In this paper a high performance system with the latest NBISOM_25 chips is presented. The NBISOM_25 integrated circuit contains 25 processing elements in a 5 by 5 array. Due to the scalability of the chips a VME-bus board was built with 16 ICs on it. The controller for the VME-bus and the SOFM hardware are realized using FPGAs. The system runs SOFM applications with up to 400 elements in parallel mode (20 by 20 map). Each weight vector can have up to 64 weights of 8 bit accuracy. The maximum performance of the board-system is 4.1 GCPS (recall) and 2.4 GCUPS (learning).
A High Performance SOFM Hardware%2DSystem.
Many applications of Selforganizing Feature Maps (SOFMs) need a high performance hardware system in order to be efficient. Because of the regular and modular structure of SOFMs, a hardware realization is obvious. Based on the idea of a massively parallel system, several chips have been designed, manufactured and tested by the authors. In this paper a high performance system with the latest NBISOM_25 chips is presented. The NBISOM_25 integrated circuit contains 25 processing elements in a 5 by 5 array. Due to the scalability of the chips a VME-bus board was built with 16 ICs on it. The controller for the VME-bus and the SOFM hardware are realized using FPGAs. The system runs SOFM applications with up to 400 elements in parallel mode (20 by 20 map). Each weight vector can have up to 64 weights of 8 bit accuracy. The maximum performance of the board-system is 4.1 GCPS (recall) and 2.4 GCUPS (learning).
Hardware Design for Self%2DOrganizing Feature Maps with Binary Input Vectors.
:  A number of applications of self organizing feature maps require a powerful hardware. The algorithm of SOFMs contains multiplications, which need a large chip area for fast implementation in hardware. In this paper a resticted class of self organizing feature maps is investigated. Hardware aspects are the fundamental ideas for the restictions, so that the necessary chip area for each processor element in the map can be much smaller then before and more elements per chip can work in parallel. Binary input vectors, Manhatten Distance and a special treatment of the adaptation factor allow an efficient implementation. A hardware design using this algorithm is presented. VHDL simulations show a performance of 25600 MCPS (Million Connections Per Second) during the recall phase and 1500 MCUPS (Million Connections Updates Per Second) during the learning phase for a 50 by 50 map. A first standard cell layout containing 16 processor elements and full custom designs for the most important part...
