Past ISSNIP Weekly Seminars and Group Meetings

2006
Date	Time	Venue	Speaker	Title
7/7/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	M Palaniswami swami at ee.unimelb.edu.au	ARC Research Network on ISSNIP
14/7/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Sophie Kaplantzis sophia.kaplantzis at eng.monash.edu.au	Security in Sensor Networks
21/7/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Ahsan Khandoker a.khandoker at ee.unimelb.edu.au	Sleep Apnea and Heart Rate Variability
28/7/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Bharat Sundaram bharats at ee.unimelb.edu.au	Micro UAVs and Geolocation
4/8/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Daniel Lai d.lai at ee.unimelb.edu.au	An Inexact Penalty Method for the Semi Parametric Support Vector Machine
11/8/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Alistair Shilton apsh at ee.unimelb.edu.au	Support Vector Machines - Classification, Regression and Clustering: A Unified Approach
		Abstract: Since their inception in the mid 90s, support vector machines have become popular tools for a wide variety of tasks, ranging from pattern classification, regression, clustering and density estimation, to name a few. The aim of this talk is to give a brief introduction to the subject, concentrating on the underlying geometry of the problem.  The talk will cover pattern classification, regression and clustering, and the connection between these formulations.
18/8/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Adil Bagirov Centre for Informatics and Applied Optimization, School of Information Technology and Mathematical Sciences, University of Ballarat. a.bagirov at ballarat.edu.au	Max-min separability and its application in data classification
		Abstract: In this talk we discuss the problem of discriminating two finite point sets in the /n/-dimensional space by a finite number of hyperplanes generating a piecewise linear function. If the intersection of these sets is empty then they can be strictly separated by a max-min of linear functions. We introduce an error function and discuss an algorithm for its minimization. This function is non-convex piecewise linear. An algorithm incrementally calculating hyperplanes separating two sets is described. We consider the application of this algorithm for solving supervised data classification problems. The results of numerical experiments using real-world datasets will be presented.
25/8/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Jayavardhana Gubbi jrgl at ee.unimelb.edu.au	Protein Secondary Structure Prediction
		Abstract: Protein structure prediction is a powerful tool in today_s drug esign industry. Talk will introduce the problem of protein structure prediction particularly prediction of secondary structure. We present some new results in protein secondary strucutre prediction using support vector machines. This will be based on the combination of Chou-Fasman parameters, physico-chemical parameters and position specific scoring matrix. This will be a short talk for about 20 mins (Conference Practice Talk).
01/9/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Wei Hua Wang weihw at ee.unimelb.edu.au	Network flow control for real-time services
		Abstract: This talk is concerned with flow control and resource allocation problems in computer networks in which real-time applications may have hard Quality of Service (QoS) requirements. Recent optimal flow control approaches are unable to deal with these problems since QoS utility functions generally do not satisfy the strict concavity condition in real-time applications. For elastic traffic, we show that bandwidth allocations using the existing optimal flow control strategy can be quite unfair. If we consider different QoS requirements among network users, it may be undesirable to allocate bandwidth simply according to the traditional max-min fairness or proportional fairness. Instead, a network should have the ability to allocate bandwidth resources to various users, addressing their real utility requirements. For these reasons, we proposes a new distributed flow control algorithm for multi-service networks, where the application's utility is only assumed to be continuously increasing over the available bandwidth. In this, we show that the algorithm converges, and that at convergence, the utility achieved by each application is well balanced in a proportionally (or max-min) fair manner.
8/9/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Thomas Hanselmann hanselmann at ee.unimelb.edu.au	Bayesian Belief Networks
22/9/2006		RMIT University	Machine Learning Workshop (updated) (Contact Daniel Lai for further information) d.lai at ee.unimelb.edu.au
29/9/2006	11am	CSSIP Board Room Dept of EEE Building 193, Level 3 The University of Melbourne	Dr. Slaven Marusic s.marusic at ee.unimelb.edu.au	Wavelet based feature enhancement
		Abstract: The seminar will provide an overview of the discrete wavelet transform with a particular emphasis on its use in feature enhancement applications. Issues concerning the potential advantages of a multiresolution based representation of input data and the ability to exploit such properties for use as a prepocessing stage for signal enhancement and classification will be considered. Applications including image denoising and feature detection in medical imaging will also be discussed.
27/10/2006	11am	Green Theatre Dept of EEE Building 193 The University of Melbourne	Prof. David Suter Monash University, Australia	Measurement Fusion by Rank/Subspace Constraints
		Abstract: Repeated measurements of a quantity (directly or indirectly) puts extra constraints on the quantity of interest. If there are more constraints than the degrees of freedom, then the "extra" constraints can be used to reduce the effects of noise (and/or missing measurements). In the case of linear subspace (rank) constraints, estimates of the amount of "denoising" can be obtained. This talk will illustrate with several examples and also discuss methods of exploiting such constraints. The examples will be drawn from computer vision but a wide variety of applications share the same fundamental structure and thus the ideas should apply quite generally
3/11/2006	11am	Green Theatre Dept of EEE Building 193 The University of Melbourne	Jin Shong Sun j.sun at ee.unimelb.edu.au	Active Queue Management--a mechanism to Improve performance of congestion control
		Abstract: This talk is concerned with active queue management (AQM) problems in computer networks. We discuss the motivation and objectives of active queue management, survey the development of AQM, and discuss some open problems about AQM.
01/12/2006	11am	Brown Theatre Dept of EEE Building 193 The University of Melbourne	A/Prof M Palaniswami swami (at) ee.unimelb.edu.au	Convergence of Intelligent Sensors, Sensor Networks and Information Processing
		Abstract: The convergence of smart sensors, sensor networks and information processing creates exciting opportunities to solve some of the demanding and complex problems of modern society. This interdisciplinary field also brings together several disciplines in electrical engineering, computer science, mechancial engineering, Geomatics, Civil engineering and Mathematics. This seminar will present some cross section projects in areas of healthcare, critical infrastructure, defence and environment. The sample projects include snesor networks for the Aged (healthcare), montoring the bridges and roads (infrastructure), unmanned aerial vehicles (defence), Monitoring the Great barrier Reef (Environment). These are funded by ARC, DEST and Industry. There will be opportunities for discussion on collaboration at the local and international level.