Computing Seminar Series
Forthcoming Department of Computing Seminars
See Previous Seminars
DSP Technology: How did it become so powerful, and what can we expect in the future?
Speaker: Dr. Ronald W. Schafer
Dates: 24/04/12
Time: 11:00-13:30
Location: AG03
Overview: The field of digital signal processing technology has a rich history and many accomplishments that are a source of pride and satisfaction to the many engineers and researchers who have contributed over the past 50 years or more. This talk will begin by considering some of the most significant of past results and some of the key contributors to these achievements. Our purpose will be to highlight things that can be learned from these accomplishments that could be useful as we go forward. While DSP has become ubiquitous in modern technology-driven life, the field is far from nearing the end of its development. The past has merely provided a substantial platform from which to reach for even greater accomplishments. This will be demonstrated by a discussion of current topics in DSP education and research and by speculation about what might be achieved in potential new applications of the continuously evolving technology of DSP.
Speaker Bio: Ronald W. Schafer received BSEE (1961) and MSEE (1962) degrees from the University of Nebraska and a Ph.D. (1968) degree from MIT. From 1968 to
1974 he was a member of the Acoustics Research Department, Bell Laboratories, Murray Hill, NJ, where he contributed to some of the earliest research on digital signal processing. In 1974 he joined Georgia Tech as John and Marilu McCarty Professor of Electrical and Computer Engineering.
Over a thirty-year career at Georgia Tech, he introduced literally thousands of students to the field of digital signal processing and supervised graduate student research in speech processing, image processing, biomedical signal processing, and communication signal processing. He retired from Georgia Tech as Professor Emeritus in 2004. Now he is a HP Fellow in the Mobile & Immersive Experience Laboratory at Hewlett-Packard Laboratories in Palo Alto, CA, where his research focuses on acoustic and sensor signal processing. He continues teaching DSP courses as a Consulting Professor at Stanford University.
Dr. Schafer is a member of the National Academy of Engineering and a Fellow of both the IEEE and the Acoustical Society of America. He has co-authored eight widely used textbooks in the DSP field including, most recently, Signal Processing First (2003), Discrete-Time Signal Processing, 3e (2010), and Theory and Application of Digital Speech Processing (2011). He has received numerous awards for his teaching and research including the 1980 IEEE Emanuel R Piori Award, the 1985 Distinguished Professor Award from Georgia Tech, the 1992 IEEE James H. Mulligan, Jr. Education Medal, and the 2010 IEEE Jack S. Kilby Signal Processing Medal.
Previous Department of Computing Seminars
- Multi-objective Evolutionary Algorithms: Applications and Technology
- Optimization in Computational Finance and Economics
- The Utility of Human Social Motivations: Biology, Individuality, and Culture
- Detecting Birdsongs via Periodic Structures: Robust Signal Processing Methods for Applied Nature Conservation
- Domain specific languages: why? how? and where next?
- Reinforcement Learning by value-gradients, and Neuro-pilot demos
- Computing in Healthcare: Some applications of data analysis in the medical domain
- Parametricity and Dependent Types; (joint work with Jean-Philippe Bernardy and Patrik Jansson)
- Elaborating Operational Requirements from Goal Models.
- Applying computer hacker methodology to conduct surveillance on and ultimately disrupt criminal and terrorist social networks.
Multi-objective Evolutionary Algorithms: Applications and Technology
Speaker: Prof Lyndon While (Walking Fish Group,The University of Western Australia)
Dates: 18/11/11
Time: 14:00-15:30
Location: AG08
Overview: Evolutionary algorithms have been applied successfully to a wide range of multi-objective optimisation problems, but the technology is still maturing. In this talk I will discuss some of the problems to which the WFG has applied MOEAs, and some of the contributions that we have made to the underlying technology.
Speaker Bio: Lyndon While leads the Walking Fish Group in Perth, Western Australia, which performs research using MOEAs in collaboration with local industry and other academics around the world.
Optimization in Computational Finance and Economics
Date: 08/03/11
Time: 14:30-16:00
Location: C337
Speaker: Edward Tsang (University of Essex)
Overview: Optimization as a problem where given an objective (or objectives), the goal is to to find optimal (i.e. the best) or near-optimal (i.e. close to the best), solutions. Economists and financial experts focus on the measure of success, i.e. the definition of the objectives. Computer scientists focus on the optimization techniques, i.e. *how* to find the optimal or near-optimal solutions. In fact, by knowing the optimization techniques better, economists and financial experts can formulate the problem better. By knowing what is required, computer scientists could be developing specialized techniques useful for economists and financial experts. There is synergy for the two communities to work together. In this talk, I shall be giving three examples of optimization in computational finance and economics, namely, portfolio optimization, automated bargaining and economic wind-tunnels.
Speaker Bio: Edward Tsang holds a first degree in Business Administration (1977, major in Finance) and a PhD degree (1987) in Computer Science. Prior to his PhD studies, he served for five years in various positions in the commercial sector in Hong Kong. He is currently a Professor in Computer Science at University of Essex. Edward Tsang was a co-founder of Centre for Computational Finance and Economic Agents (CCFEA). CCFEA is an interdisciplinary research centre, which applies artificial intelligence methods to problems in finance and economics. It is supported by City Associates, which is supported by HSBC, Olsen & Associates, Ionic Sharescope, Bank of England, Old Mutual and other companies. He is the author of Foundations of Constraint Satisfaction, the first book to define the scope of the field, and arguably the most rigorous in the field (all major concepts were defined with formal logic). He founded the Technical Committee in Computation Finance and Economics in IEEE's Computational Intelligence Society in 2004, and chaired it until the end of 2005. He serves in editorial boards of IEEE Transactions in Evolutionary Computation, Constraints, the Scheduling journal and other journals. He has been a member of the Computing College of Engineering and Physical Sciences Research Council (EPSRC, UK) since 1997. He has served committees and panels for many major international conferences and workshops. He has given invited speeches in international conferences, including the Parallel Problem Solving from Nature (PPSN) 2006. Edward Tsang's research is highly industry-relevant. He has given consultation to GEC Marconi, British Telecom, the Commonwealth Secretariat, Old Mutual Asset Managers, RAS and other organizations. Guided Local Search, developed in his laboratory, has been embedded in ILOG DISPATCHER, a commercial product for vehicle routing. EDDIE, a financial forecasting tool, has attracted much commercial interest.
The Utility of Human Social Motivations: Biology, Individuality, and Culture
Date: 28/02/11
Time: 14:30-16:00
Location: C302
Speaker: Joanna Bryson (Computer Science Department at University of Bath)
Overview: In the 1990s, John Laird (the caretaker for and one of the founders of SOAR) declared that there were no applications for human-like AI other than military simulations, and possibly computer games. Cognition is a relatively unpopular strategy in nature, exploited by relatively few species. Yet as we enter the second decade of the 21st century, interest in (and funding for) cognitive and social AI systems has never been greater. In this talk I introduce AI simulations to examine the biological origins and mechanisms of human-like intelligence. Humanity is striking not only for being cognitive, but also for being social and altruistic. The costs of social living are obvious: competition for the best resources. Traditionally, biology has considered the primary benefit of social living to be security, both from predators and possibly from competing groups. But recent evidence from field studies, experiments & theory allows a more radical hypothesis. Perhaps the altruistic sharing of information and behaviour -- what we call culture in humans -- is an essential strategy for any species that exploits cognitive strategies. In this talk I will review a variety of evidence supporting the universal exploitation of socially-acquired behaviour by cognitive species, including field work (observation of nature), laboratory experiments and emphasising my own work in social simulation. I will then move on to look at work in progress: first examining the role of individuality within the collective, and second of subcultural variation in balancing the opposing needs for innovation and knowledge retention in human culture. I will then briefly discuss the implications of these outcomes on how AI should be built and how cognitive systems can be incorporated into human society.
Speaker Bio: Joanna J. Bryson specialises in two areas: the advancement of systems artificial intelligence (AI), and the use of AI simulations to further the understanding of natural intelligence, including human culture. She holds degrees in behavioural science, psychology and artificial intelligence from Chicago (BA), Edinburgh (MSc and MPhil), and MIT (PhD). She joined The University of Bath in 2002, and was made a Reader in Computer Science in 2010. Between 2007-2009 she held the Hans Przibram Fellowship for EvoDevo at the Konrad Lorenz Institute for Evolution and Cognition Research in Altenberg, Austria. She is currently a visiting research fellow in the University of Oxford's Department of Anthropology, where she is working on Explaining Religion. At Bath she heads Artificial Models of Natural Intelligence, where she and her colleagues publish in biology, anthropology, cognitive science and systems AI.
Detecting Birdsongs via Periodic Structures: Robust Signal Processing Methods for Applied Nature Conservation
Date: 22/02/11
Time: 14:30-16:00
Location: C337
Speaker: Daniel Wollf (Music Informatics Group, Department of Computing, City UNiversity London)
Overview: The talk will be about a collaborative bioacoustics project of the Multimedia Signal Processing Group of Bonn University and the Animal Sound Archive in Berlin. Long term monitoring recordings have been performed in a nature conservation area at Parstein Lake, Brandenburg, Germany. Besides a general description of the Project, the talk will focus on signal processing algorithms which are currently used to identify the Savi's Warbler calls within such recordings. The design of the algorithms heavily orients towards robust recognition of the species' repetitive song, as the undirected recordings are subject to varying grades of noise. For more background, find attached the link to my master's thesis: http://juppiemusic.com/wp-content/uploads/2010/07/Daniel_Wolff_Bird_Detection.pdf and to Rolf's PhD thesis "Algorithmic Analysis of Complex Audio Scenes" which also evolved during the project: http://hss.ulb.uni-bonn.de/diss_online/math_nat_fak/2008/bardeli_rolf
Speaker Bio: Born in 1983 in Cologne, Germany, Daniel studied computer science and musicology in Bonn, working as a student assistant in the Multimedia Signal Processing Group of Prof. Clausen. Central topics in his work then included music information retrieval and bioacoustic pattern recognition. After finishing his CS degree, parts of which have been published in the Elsevier Pattern Recognition series, Daniel continued his studies in musicology at the Bonn University Department of Sound Studies, in 2008, focussing on cultural influences in algorithm design for music retrieval. Granted an PhD scholarship from City University, Daniel started his current research in 2010 at the Music Informatics Research Group.
Domain specific languages: why? how? and where next?
Date: 09/12/2010
Time: 14:30-16:00
Location: C161
Speaker: Laurence Tratt (Middlesex University)
Overview: While the concept of Domain Specific Languages (DSLs) continues to gain in popularity and importance, there is surprisingly little understanding about what they are, when they're applicable, their benefits, and their costs. In this talk I will first motivate the need for DSLs, and explain when and where they might be applicable. I will then consider some of the different ways they can be implemented, particularly by embedding them into programming languages such as the syntactically extendable language Converge (http://convergepl.org/) that I have developed which, at a simple level, can be thought of as Python with Template Haskell-esque macros. Finally I will look forward: what might the next stage in the development of DSL technologies be?
Speaker Bio: Laurence Tratt is a Senior Lecturer at Middlesex University and software consultant; previously he was a Senior Lecturer at Bournemouth University. His research interests include domain specific languages and modelling, where he has contributed to several international standards. He is the chief designer and implementer of the Converge programming language. He is an Associate Editor in Chief of IEEE Software and sits on the Editorial Board of the The Journal of Object Technology. He can be reached via http://tratt.net/laurie/
Reinforcement Learning by value-gradients, and Neuro-pilot demos
Date: 18/11/2010
Time: 10:30-12:00
Location: AG11
Speaker: Mike Fairbanks (Department of Computing at City University)
Overview: Reinforcement learning often makes use of a value function that can be represented by a neural network. Finding efficient or stable algorithms to train the neural network is a difficult problem. We present a method of learning the value function by considering its gradient with respect to the state vector. This leads to increased efficiency and a convergence proof for the learning algorithm. The method has been applied successfully to a simulated lunar-lander spacecraft in a 2d environment, which will be demonstrated here.
Speaker Bio: Prior to starting a part-time PhD at City university, Mike Fairbank has been an independent researcher in neural networks and reinforcement learning for 10 years. Most of his research has been driven by the problem of making a simulated spacecraft learn for itself to navigate through increasingly complex environments. This has led to insights into value-function learning, and neural network training algorithms. In 2008-2009 Mike was a panellist for the US' National Science Foundation awarding funding to major cybernetics research projects. Mike studied Mathematical Physics at Nottingham University, followed by an MSc in Knowledge based systems at Edinburgh. His full time employment is maths teaching in a sixth form college in London.
Computing in Healthcare: Some applications of data analysis in the medical domain
Date: 10/11/2010
Time: 14:00-15:30
Location: AG03
Speaker: Peter Weller(Centre for Health Informatics at City University London)
Overview: The increasing use of technology in healthcare has resulted in it becoming a data rich area. Large quantities of data are now routinely collected for many treatments and procedures. Yet, apart from immediate patient care and statistical returns, very little long term use is made of these datasets. However, with processing and the use of analytical tools, additional information can be extracted from the data. This knowledge can provide the clinicians with additional insight into a person's condition and so improve patient care. This seminar will present some ongoing work in this area. A number of real world problems will be introduced and current work to address them described.
Speaker Bio: Peter Weller is a Senior Lecturer in Medical Informatics in the Centre for Health Informatics. Prior to an academic career he spent over ten years in industry designing medical equipment. His last industrial appointment was as Manager of a multi-disciplined research and development department. Current areas of research interest are mainly based in the high dependency environment and include: monitoring of complex, critical systems; wearable computer technology; biomedical signal processing; clinical decision support systems; applications of artificial intelligence and non-linear systems for clinical knowledge management, mathematical modelling of cardiac function; and robotic surgery. Current professional activities include serving on the Executive Team of the Healthcare Technologies Professional Network of the IET and the editorial board of Critical Reviews in Biomedical Engineering. He has held honorary research positions at St. Bartholomew's and Royal Brompton Hospitals, and was a visiting Professor in Department of Anaesthesia, University of British Colombia, Canada.
Parametricity and Dependent Types; (joint work with Jean-Philippe Bernardy and Patrik Jansson)
Date: 10/10/2010
Time: 14:00-15:30
Location: AG03
Speaker: Ross Paterson(Department of Computing at City University London)
Overview: John Reynolds showed that polymorphic terms satisfy abstraction properties corresponding to their types, a result dubbed "Theorems for Free" by Phil Wadler. This result has been applied to such areas as automated theorem proving, test case generation and program optimization. We address the extension of this result from the second-order lambda calculus to pure type systems, in which a wide variety of powerful type systems may be expressed. Exploiting the "Propositions as Types" idea leads to a particularly simple form in that more general setting.
Elaborating Operational Requirements from Goal Models.
Date: 28/04/2010
Speaker: Alessandra Russo (Department of Computing at Imperial College London)
Overview: In Goal Oriented Requirements Engineering, a key activity is the elaboration of correct and complete operational requirements, in the form of pre- and trigger-conditions, that satisfy given system goals. The few existing approaches that provide support for this crucial task mainly rely on significant effort and expertise of the engineer. In this talk, we present a tool-based approach that combines model checking, inductive learning and scenarios to elaborate operational requirements from goal models. This is an iterative process consisting of four phases: (i) the analysis of existing partial requirements with respect to a goal model, (ii) the elaboration of positive and negative scenarios from counterexamples, (iii) the computation of (alternative) missing operational requirements that cover all positive scenario and eliminates all negative ones, and (iv) the selection of operational requirements from suggested alternatives. We will focus in particular on the third phase and show how formal requirements can be automatically inferred from positive and negative scenario descriptions, whilst preserving consistency with the system goals. This phase is based on translating the partial requirements, goals and scenarios into an event-based logic programming formalism and using a non-monotonic Hybrid Abductive Inductive Learning system to learn a set of missing pre- and trigger-conditions. The contribution of this work is a novel application of ILP to Requirements Engineering that also demonstrates the need for non-monotonic learning.
Speaker Bio: Dr. Alessandra Russo is a Senior Lecturer in the Department of Computing at Imperial College London. where she obtained her PhD in mathematical logic. She has substantial research experience in the application of Computational Logic to Software Engineering. In the last ten years, she has played a central role in the development of formal reasoning systems for software engineering, which have gained international recognition. She has been part of the IFIP Working Group 2.9 on Requirements Engineering, given invited talks at many UK and abroad academic institutions, and been one of the invited speakers at the International Workshop on Formal Approaches to Ubiquitous Systems (2009) together with Professor Marta Kwiatkowska and Professor Robin Milner. Dr Russo is author of over 80 publications in international conferences and journals, two of which were awarded prize for best application paper. She has served on the program committees of many international conferences, and is currently investigator on various EPSRC and US funded research projects. Since March 2005, she has also been Editor-in-Chief of the IET Software journal.
Applying computer hacker methodology to conduct surveillance on and ultimately disrupt criminal and terrorist social networks.
Date: 17/03/2010
Speaker: Stewart Bretam
Overview: In the information age the Internet has become one of the dominating forms of human communication in both Western society and the developing world with new social networking and micro blogging sites emerging almost daily. This combined with the fact that the old social hierarchy of monarch, government and institution is now becoming defunct in favour of flatter networks of people, presents a unique opportunity for agencies engaged in information gathering and intelligence production.
By the application of a methodology traditionally reserved for computer hackers and willingness to exploit what Lyon terms as 'self-panopticonism,' there is a unique opportunity to 'hack' social networks engaged in terrorist and criminal activity. This paper examines the potential application for such a methodology within the context of hacker terminology.
Starting with the process of 'foot printing' there is an examination of the types of information that is available on the Internet and how these can be combined together according to the 'mosaic philosophy' to form a coherent intelligence picture. From this initial starting point the equivalent of 'network penetration' is examined with ways that the wider network can be exposed by exploiting weaker individuals within the system using 'social engineering.' Finally there is a brief theoretical examination of how the equivalent of a 'denial of service' attack can be performed on a social network.
The breadth and depth of this data is often surprising, bank account details, social security numbers, age, number of children, place of work, spending habits and martial status are but a tiny fraction of the data that is now freely available on the internet regarding a huge proportion of the world's population.
Speaker Bio: Stewart Bertram, a former military intelligence analyst with experience in both Iraq and Afghanistan, now currently working within the Center for Investigative, Security and Policing science at City University in London.