Cancelled due to speaker circumstances.
Date: Wednesday 16th December 2009, 11:15, CS103 Computer Science, Univ. of York
Speaker: Sandy Louchart, Heriot-Watt University, School of Mathematics and Computer Science
Date: Thursday 10th December 2009, 14:00, CS202J Computer Science, Univ. of York
Speaker: Sam Delvin, University of York, Department of Computer Science, Artificial Intelligence Group
Re-reinforcement Learning for Multi-Agent Systems
Research into the application of Reinforcement Learning to Multi-Agent Systems has resulted in a vast quantity of new algorithms, many proven to converge under very specific conditions. This seminar will summarise the work done, highlighting the key algorithms and difficulties of application. A number of techniques for overcoming such problems are introduced, some of which begin to identify areas of potential future research.
Date: Wednesday 2nd December 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Bo Wang, University of York, Department of Mathematics
Gaussian process modelling for functional data
Gaussian processes modelling has been shown to be a practical, flexible and powerful tool in nonlinear data analysis and has received increasing interests in various fields, such as statistics, machine learning and biology. In this talk, I will first give a brief introduction on Gaussian processes regression and classification, then will present one of its recent developments in analysis of functional data.
Date: Wednesday 25th November 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Sach Mukherjee, Warwick University, Statistics and Complexity
Network learning in cancer biology
Networks of proteins called "signalling networks" play a key role in the control of diverse cellular processes; their aberrant functioning is heavily implicated in many diseases, including cancer. Aberrations in cancer cells are thought to perturb normal connections in these networks, with important biological implications. Yet tumour-specific signalling network connectivity remains poorly understood, especially at the level of relevant (post-translational) protein modifications. However, high-throughput biochemical technologies are now able to make measurements on components of these systems, and can, in principle, shed light on a variety of open questions concerning signalling in cancer. I will discuss machine learning approaches for interpreting these data, in particular how probabilistic graphical models can be used to integrate biochemical data and prior knowledge of signalling biology to facilitate the discovery process.
Date: Wednesday 18th November 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Martin Brain, University of Bath, Computer Science
Future Directions in Logic Programming
This talk covers one of the current 'hot' areas in logic programming, answer set programming. We start with an introduction to answer set semantics and where it fits in the context of non-monotonic logics and "common sense" reasoning provides. This provides a foundation for discussing answer set programming and it's relation to CSP and SAT based paradigmes. Finally we discuss some of the application development work that has been done at the University of Bath and explain how all of this relates to composing music, playing computer games and proving machine code to be optimal.
Date: Wednesday 4th November 2009, 11:15, CS103 Computer Science, Univ. of York
Speaker: David Grace, University of York, Department of Electronics
Cognitive Communications - The Intelligent Future?
Wireless communications systems today largely rely on fixed rules and limited adaptivity, especially when it comes to using the radio spectrum and selecting network path selection. Such methods are increasingly leading to bottlenecks in supply, and in some cases stifling innovation. A new research approach, Cognitive Communications, which encompasses the rapidly expanding fields of cognitive radio and cognitive networking, has the potential to revolutionise future communications systems by collectively applying intelligence, adaptivity and flexibility to resource assignment and usage. This talk will discuss how true 'cognitive' communications can be realised by combining the disciplines of wireless communications and distributed artificial intelligence, to create wireless communication systems that are capable of learning about their environment and dynamically adapting their behaviour through reasoning. The findings of a recently completed large research project, 'Cognitive Routing for Tactical Ad Hoc Communications' will be used to support this perspective.
Date: Wednesday 21st October 2009, 11:15, CS103 Computer Science, Univ. of York
Speaker: Shuguang Li, University of York, Department of Computer Science
Title: Information-seeking Question Generation in Interactive Question Answering
Question Answering systems have received a lot of interest from NLP researchers during the past years. But it is often the case that traditional QA systems cannot satisfy the information needs of the users as the question processing part may fail to properly classify the question or the information needed for extracting and generating the answer is either implicit or not present in the question. In such cases, interactive information-seeking dialogue is needed to clarify the information needs and reformulate the question in a way that helps an interactive QA system to find the correct answer. The topic of a good information-seeking question should not only be relevant to the user's question but should distinguish each answer from the others so that the new information can reduce the ambiguity and vagueness in the user's question. We proposed a cluster ranking algorithm to generate topics for information-seeking questions.
Date: Wednesday 1st July 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: James Cussens, The University of York, Computer Science
Title: A tutorial on logic-based approaches to SRL
The relations in Statistical Relational Learning are often expressed using first-order logic, leading to formalisms which combine both logical and probabilistic representations. In this talk I intend to explain the most important consequences of adopting a logical approach to SRL. Defining distributions over 'possible worlds' is a common theme to many such approaches. Two prominent logic-based formalisms - Markov logic networks and PRISM programs - will be used as exemplars. Although the talk is tutorial in nature, I hope to make it interesting to those already familiar with this area!
Date: Wednesday 10th June 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Amer Al-zaidi, The University of York, Computer Science
Title: Islamic Finance and A Potential for Innovation
Credit crunch is pushing for an alternative mode of financing. It is the chance to market Islamic finance globally. Islamic finance needs more development in many areas such as Management, accounting, finance and IT. Islamic finance is the fastest growing area. This presentation will cover the basic concepts about Islamic finance and illustrates the most important products. Multi-agent systems and equation discovery can offer a lot to Islamic finance which this presentation going to describe how Islamic finance can adapt these techniques.
Date: Wednesday 3rd June 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Yu Xiong
Dr. Yu Xiong - is a Research Fellow in Management Systems, specialising in system modelling. His main research interests are operations management and its applications.
Title: Structuring New Product Development (NPD) Pipelines Considering Resource Constraints
New Product Development (NPD) is a crucial process to keep pharmaceutical companies competitive. However, because of its inherent features, this process is laden with high risk and uncertainly, and it is well known that no single development approach will necessarily lead to a successful product. In an attempt to manage this risk, multiple projects with multiple stages are generally funded within pharmaceutical companies. In such a resource constrained environment, senior managers have to grapple with the challenge of setting levels of funding for a portfolio of projects, while operational managers have to choose the right number of approaches to be funded at each stage of the NPD process. This paper describes a mathematical model, which is dependent on the magnitude of the business opportunity, cost per development approach and survival probabilities of candidates; and proposes solutions to structure NPD pipelines from a portfolio of projects in a resource constrained environment. Using a real-world scenario from the pharmaceutical industry, we explain the theoretical and practical contribution of this model.
Date: Wednesday 20th May 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Asim Karim, The University of York, Computer Science
Dr. Asim Karim is a visiting academic researcher from Lahore University of Management Sciences (LUMS), Pakistan, where he is an associate professor of computer science. His research interests are machine learning, data mining, and applications. In recent times, he has focussed on textual data and Web applications.
Title: Document Classification and Clustering: A Discriminative Information Pooling Approach
The vast majority of information available today exists in the form of text documents. Some common examples include e-mails, Web pages, research publications, and social computing contents. The organization and understanding of such information requires efficient and robust algorithms for classification and clustering. In this talk, I will present a simple approach based on the discrimination information provided by the terms in the documents. The collective discriminative information provided by all terms in a document, or the discrimination information pool, provides the key measure for classifying and clustering text documents. I will also present some results illustrating the effectiveness of the approach.
Date: Wednesday 6th May 2009, 11:15, CS202J Computer Science, Univ. of York
M.Eng. Student Talks.
Speaker: Rob Taylor, Computer Science Department, The University of York.
Title: Evaluation and Comparison of a New Edge Reversal Move for Markov Chain Monte Carlo for Bayesian Networks
Background: Improving the rate at which Markov Chain Monte Carlo (MCMC) samplers converge to the desired posterior distribution is an ongoing research problem in Machine Learning. Classic standard structure MCMC methods are slow at converging and have recently been surpassed in performance by node ordering MCMC samplers. These node orders are too restrictive and implausible when no knowledge of the domain is known prior to the structure learning and thus improvements upon standard structure MCMC are required. Recent research has shown, by modifying the classical reverse edge move of the standard structure MCMC sampler to sample new parents for both nodes on the edge, it is possible to drastically speed up the rate of the MCMC sampler. This new edge reversal move is presented and implemented and then several evaluations on it are performed, including a comparison with the classic standard structure MCMC sampler.
Speaker: Matthew Patrick, Computer Science Department, The University of York.
Title: Online Evolution for Unreal Tournament 2004
Background: Few commercial computer games employ machine learning techniques because the way agents appear to behave is more important to the player than how they actually work. However, limitations in the artificial intelligence of otherwise sophisticated computer games may cause players to see increased realism in graphics and experience as uncanny rather than engaging. As adaptation can be considered a necessary part of intelligence, including it in a game could improve its believability. This project combines academic concepts about machine learning into a practical approach for adapting agent tactics within a modern computer game. Efficient mechanisms are described for online evolution with games featuring both one-on-one competition and more complex interaction. In both cases, the agents learnt tactics capable of defeating pre-set static configurations, but learning is faster when more agents are evolved together. The results have been analysed to provide insight into which techniques are effective in each situation and allow an understanding of their usefulness within a commercial computer game.
Speaker: Sam Devlin, Computer Science Department, The University of York.
Title: Reinforcement Learning of Robotic Soccer under Partial Observability
Background: Existing research has approached partial observability in reinforcement learning through the use of manually coded rules for execution of information gathering actions in the policy using heuristic knowledge. However, such a solution requires explicit expert knowledge of the problem domain. By eliminating the hand-coded rules a more flexible solution can be devised that automatically learns which actions gather information and when to use them.
Date: Wednesday 18th Computer Science, Univ. of York
11:15, CS202J Computer Science, Univ. of York
Speaker: James Cussens, The University of York, Computer Science
Title: An introduction to Hidden Markov Models
Hidden Markov models (HMMs) are widely used in a number of fields, for example biological sequence analysis and natural language processing. This talk is a tutorial and assumes no prior knowledge of HMMs. The topics covered will be: 1) what an HMM is; 2) problems that can be solved using an HMM; 3) algorithms for solving these problems (Viterbi and forward/backward probabilities) and 4) methods for estimating the parameters of an HMM from data. Example HMMs and demos will be used in the tutorial.
14:00, CS103 Computer Science, Univ. of York
Speaker: Alan Frisch, The University of York, Computer Science
Title: The Proper Treatment of Undefinedness in Constraint (and perhaps other) Languages
Implementations of constraint languages handle undefined expressions in an unsystematic and sometimes inconsistent manner. This talk addresses the systematic treatment of undefined expressions in constraint languages. The paper first presents three alternative semantics for a simple constraint language that has undefined expressions. On constraint models that contain no undefined expressions the three semantics agree with each other and, we believe, with the intuitions of constraint language users. Importantly, the constraint solving technology that is used to solve problems expressed in a constraint language does not support any notion of undefinedness (except perhaps raising an exception). The talk shows, for each of the semantics, how any constraint model that contains undefined expressions can be translated to one that doesn't but still has the same set of solutions. As undefinedness does not arise in these resulting models, they can be implemented correctly by existing constraint solvers.
Date: Wednesday 11th March 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Alan Black, Carnegie Mellon University, Language Technologies Institute
Title: Speech Synthesis: past, present and future and how it mirrors speech processing development in general
Biography: Alan W Black is an Associate Professor in the Language Technologies Institute at Carnegie Mellon University. He previously worked in the Centre for Speech Technology Research at the University of Edinburgh, and before that at ATR in Japan. He is one of the principal authors of the free software Festival Speech Synthesis System, the FestVox voice building tools and CMU Flite, a small footprint speech synthesis engine. He received his PhD in Computational Linguistics from Edinburgh University in 1993, his MSc in Knowledge Based Systems also from Edinburgh in 1986, and a BSc (Hons) in Computer Science from Coventry University in 1984. Although much of his core research focuses on speech synthesis, he also works in real-time hands-free speech-to-speech translation systems (Croatian, Arabic and Thai), spoken dialog systems, and rapid language adaptation for support of new languages. Alan W Black was an elected member of the IEEE Speech Technical Committee (2003-2007). He is currently on the board of ISCA and on the editorial board of Speech Communications. He was program chair of the ISCA Speech Synthesis Workshop 2004, and was general co-chair of Interspeech 2006 -- ICSLP. In 2004, with Prof Keiichi Tokuda, he initiated the now annual Blizzard Challenge, the largest multi-site evaluation of corpus-based speech synthesis techniques.
Abstract: This talk will look at the past, present and future of speech synthesis and how it relates to speech processing development in general. Specifically I will outline the advances in synthesis technology giving analogies to the developments in other speech and language processing fields (e.g. ASR and SMT) where knowledge-based techniques gave way to data-driven techniques, which in turn have pushed both machine learning technologies and later re-introduced techniques to include higher level knowledge in our data-driven approaches. We will give overviews of diphone, unit selection, statistical parametric synthesis, voice morphing technologies and how synthesis can be optimized for the desired task. We will also address issues of evaluation, both in isolation and when embedded in real tasks. While widening our view of speech processing we will also present the publicly used Let's Go Spoken Dialog System (and its evaluation platform Let's Go Lab), our rapid language adaptation system (CMUSPICE) allowing construction of ASR and TTS support in new languages by non-speech experts and out hands-free real-time two-way speech to speech translation system showing how system integration can cause cross technology innovation.
Date: Wednesday 4th March 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Azniah Ismail, The University of York, Computer Science
Title: Limiting the search space boundaries in bilingual lexicon extraction
/Bilingual lexicon extraction /is process of acquiring word pairs from corpora, where each extracted pair consists of a word in one language and its counterpart in another language. Our study generally involves bilingual lexicon extraction from non-parallel, same domain corpus with help from initial, small bilingual lexicon (/seed words/). In this work, we address search space problem in bilingual lexicon extraction. According to Haghighi et al (2008), the most common errors detected in their analysis on top 100 errors were from semantically related words which had strong context feature correlations (for example, /liberal /and /partido/). Whereas Fung and McKeown (1997) among others mentioned that most of their correct translations found among top candidates but not top one. Our technique helps by limiting the context boundary of a search space while maintaining the typical search space reduction between the input source and target within the context. The words or translations are filtered out to a manageable size using some quantitative measure, especially those false candidates that are not highly correlated with the source, in the context.
Date: Wednesday 18th February 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Mark Bartlett, The University of York, Computer Science
Title: Using Machine Learning in Worst-Case Execution Time Estimation
Many applications depend on correct timing behaviour as much as they depend on correct functional behaviour. If timing constraints are not met, the results may be even be fatal. If an aircraft does not respond to the pilot's controls in a timely manner, or car braking systems do not activate fast enough, lives could be lost. An essential part of ensuring the correct behaviour of any Real-Time System is the Worst Case Execution Time (WCET), the longest that a task can run for any possible input. Current methods for determining this value are not precise due to theoretical limitations. A new technique based on Machine Learning will therefore be presented. In particular, the sub-problem of determining the maximum number of executions of a loop will be examined. Over-estimating this quantity slightly can lead to massive over-estimations of the total time, particularly where inner loop counters are dependent on outer loop counters. The new technique is able to precisely determine the number of loop executions for even these complex cases.
Date: Wednesday 11th February 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Tommy Yuan, The University of York, Computer Science
Title: Computational Dialectics
Informal logic (IL) is an area of philosophy rich in models of communication and discourse with a heavy focus on argument and "dialogue games". Computational dialectics is a maturing strand of research that is focused on implementing these dialogue games. The aim of this talk is to introduce my two related streams of research on applying informal logic dialogue games into human-computer dialogue design. The first stream concerns research in the development of a computational system for an educational debate, and the second stream concerns the development of a computer game for abstract argumentation. The work presented has been published in several journal papers (eg International Journal of Intelligent Systems, Journal of Knowledge Engineering Review, International Journal of Artificial Intelligence in Education and Journal of Informal Logic). It is hoped that this talk can serve to generate new research synergy within the AI group.
Date: Wednesday 4th February 2009, 11:25, CS202J Computer Science, Univ. of York
Speaker: Jian Zhang, The University of York, Department of Mathematics
Title: A Bayesian model for two-way clustering
This article proposes a Bayesian method for two-way clustering and provides some statistical insights underlying this new method. We begin by embedding two-way cluster analysis into the framework of a plaid model with random effects. The corresponding likelihood is then regularised by the hierarchical priors in each layer. The resulting posterior, asymptotically equivalent to a penalised likelihood, can attenuate the effect of high dimensionality on cluster predictions. We provide an empirical Bayes algorithm for sampling posteriors, in which we estimate the layer memberships of all genes and samples by maximising an explicit marginal posterior of these layer memberships. The new algorithm makes the estimation of Bayesian plaid models computationally feasible and efficient. The performance of our procedure is evaluated on both simulated and real microarray gene expression datasets. The numerical results show that our proposal substantially out-performs the original plaid model in terms of miss-classification rates across a range of scenarios.
Date: Wednesday 28th January 2009, 11:15, CS202J Computer Science, Univ. of York
Speaker: Dimitar Kazakov, The University of York, Computer Science
Authors: Dimitar Kazakov and Tsvetomira Tsenova
Title: Equation Discovery for Macroeconomic Modelling
Keywords: Machine learning, Equation discovery, LAGRAMGE, Macroeconomic modelling, Inflation.
This article describes a machine learning based approach applied to acquiring empirical forecasting models. The approach makes use of the LAGRAMGE equation discovery tool to define a potentially very wide range of equations to be considered for the model. Importantly, the equations can vary in the number of terms and types of functors linking the variables. The parameters of each competing equation are automatically fitted to allow the tool to compare the models. The analysts using the tool can exercise their judgement twice, once when defining the equation syntax, restricting in such a way the search to a space known to contain several types of models that are based on theoretical arguments. In addition, one can use the same theoretical arguments to choose among the list of best fitting models, as these can be structurally very different while providing similar fits on the data. Here we describe experiments with macroeconomic data from the Euro area for the period 1971–2007 in which the parameters of hundreds of thousands of structurally different equations are fitted and the equations compared to produce the best models for the individual cases considered. The results show the approach is able to produce complex non-linear models with several equations showing high fidelity.