www.elsevier.com/locate/neucom

Molecular profiling technologies monitor many thousands of transcripts, proteins, metabolites or other specdI ccdIJMMIdA in a biologicF sample of interest. Givensuc high-dimensional data for different types of samples,cples,dIF6M6 methods aim to assignspecndKI to knowncwndF#J#;d Relevant feature identific##;d methods seek to de#ne a subset ofmolecTFd that differentiate the samples. This workdescFKJF LIKNON, aspecM; implementation of astatisticK approac forcrdI;D# a cdIKKIDd and identifying a small number of relevant features simultaneously. Giventwo-cMI; data, LIKNON estimates a sparse linear cneard;6 by exploiting the simple andwell-known property that minimising an L 1 norm (via linear programming) yields a sparse hyperplane. It performs well when used forretrospecJM; analysis of three creed biology pro#ling data sets, (i) small, round, blue cue tumourtransc#MF pro#les from tumour biopsies and cdd lines, (ii)sporadic breastceastdI6 transcI6T pro#les from patients with distant metastases 5 years andthose with no distant metastases 5 years and(iii) serum sample protein profiles from una#ec6DdAJDMTDdc cMTDd patients. Computationally, LIKNON is less demanding than the prevailing filter-wrapper strategy; thisapproac generates many feature subsets andequates relevant features with the subset yielding a cFJJD#dA with the lowest generalisation error. Biologic;MdA the results suggest a role for thecedI;KD micI;KDTdAFTFTT in influenc AF disease outcse and its importanc in developingcevelop decelop support systems.

This thesis focuses on applying Support Vector Machines (SVMs), an algorithm founded in the framework of statistical learning theory, to analyzing DNA microarray data. The first part

Abstract. We propose a novel online monitoring approach to distinguish between attacks and normal activity in SIP-based Voice over IP environments. We demonstrate the e ciency of the approach even when only limited data sets are used in learning phase. The solution builds on the monitoring of a set of 38 features in VoIP ows and uses Support Vector Machines for classi cation. We validate our proposal through large o ine experiments performed over a mix of real world traces from a large VoIP provider and attacks locally generated on our own testbed. Results show high accuracy of detecting SPIT and ooding attacks and promising performance for an online deployment are measured. 1