The talk will address evaluation of online classifiers that are designed to adapt to changes in data distribution over time (concept drift). A standard procedure to evaluate such classifiers is the test-then-train, which iteratively uses the incoming instances for testing and then for updating a classifier. Such learning risks to overfit, since a dataset is processed only once in a fixed sequential order while every output of the classifier depends on the instances seen so far. The problem is particularly serious when several classifiers are compared, since the same test set arranged in a different order may indicate a different winner. To reduce this risk we propose to run multiple tests with permuted data. The proposed procedure allows us to assess robustness of classifiers when changes happen unexpectedly.

Indre Zliobaite has recently joined Bournemouth University as a lecturer in computational intelligence. Her research interests include adaptive learning, detecting and handling changes (concept drift) in an online learning, data streams.

Software products are planned, developed and sold according to a typical lifecycle of a product with stages of: Introduction, Growth, Maturity and Decline. However, new software products and product lines often include disruptive innovation that results in significant changes in the lifecycle model. Wieslaw Blysz will provide you an overview of experiences, problems and solutions applied to effectively manage the challenges emerging from non-standard product lifecycle behaviour.

Wieslaw Blysz has been working is software R&D projects for 7 years, initially as a quality assurance engineer, QA team leader, department manager and currently is a CTO at software service provider company that he co-founded 4 years ago. Wieslaw Blysz obtained his MsC in Mobile Communications from Wroclaw University of Technology in Poland and accomplished Executive MBA degree.

Software for predictive analytics and data mining is no longer only used by experts in selected big companies and organisations. Advances in automation as well as a number of competitors offering solutions at an affordable price significantly extended the range of users in the last years. This presentation introduces software packages for predictive modelling and data mining that are currently available on the market. By investigating their features and differences, the current state of the art is described and potential future directions for development are shown. The work was done in the scope of the INFER project, which aims at developing solutions for next generation automated and self-adaptive predictive modelling by directly transferring latest research outcomes into user friendly software.

Christiane Lemke received a PhD from Bournemouth University in 2010 and subsequently worked there as a PostDoc. For the INFER project, she was seconded to the company REC in Wroclaw from September to December 2010.

Nowadays, the need for highly accurate adaptive modelling has become a primary aim for most state-of-the-art prediction systems. Research is currently focusing into evolving systems that maintain a high degree of autonomy, through feedback mechanisms and goal-related assistance. Genetic programming is an advance which has been established in the last two decades as a robust and effective computational intelligence method. Due to its ability to express complex domains and its evolutionary nature, genetic programming can play a significant and decisive role at most levels of such an evolving system.

Athanasios Tsakonas, Ph.D received his M.Eng in Electrical and Computer Engineering from the National Technical University of Athens and his Ph.D from University of the Aegean. Athanasios’ Ph.D thesis was 'Computational Intelligence in Complex Managerial and Financial Domains - The Evolutionary Neural Logic Network Paradigm'. Athanasios has gathered strong experience in the analysis, design and development of specialized computational intelligence systems, for the financial and medical domain. His experience includes participation in European and domestic research projects (such as BOEMIE, SHARE, EUNITE, etc.), occupation of related research positions in top research centers (such as N.C.S.R. Demokritos) or in the private sector (banks, software development companies, etc.), as well as teaching related courses in universities (Aristotle University of Thessalonika, Demokritus University of Thrace, etc.). His research interests include computational intelligence, data mining, genetic programming and complex systems. He has published 1 book and more than 45 articles, in total, in international scientific journals and conferences.

Jacek Panachida is a software engineer graduated from Wroclaw University of Technology. During his professional career in Research & Engineering Center Sp. z o.o. he was involved in implementation of several enterprise systems.

"Best practices in software development" presents the proper techniques for implementing applications and common mistakes that should be avoided.

Ludmila Kuncheva is currently a Professor at the School of Computer Science, Bangor University, UK. Her interests include pattern recognition and classification, machine learning, classifier combination and fMRI data analysis.

Wrapped in mysticism and superstition in the past, "mind reading" is now raising new scientific horizons beside ethical debates. Functional magnetic resonance imaging (fMRI) is currently the most advanced technology at the disposal of cognitive neuroscience. It measures blood oxygenation level-dependent (BOLD) signal and tries to discover how mental states are mapped onto patterns of neural activity. Feature selection and classification of fMRI data is still a formidable analytic challenge even for the state-of-the-art pattern recognition and machine learning. This talk will explain the main difficulties and approaches in fMRI data analysis. We will look at how classifier ensembles can be used for this problem. Results from an experiment will be presented, which favour the Random Oracle ensembles and Random Subspace ensembles for fMRI classification.

Trevor Martin is Professor of Artificial Intelligence at the University of Bristol. Since 2001 he has been 80% funded by BT as a Senior Research Fellow, researching soft computing in intelligent information management including areas such as the semantic web, soft concept hierarchies and user modelling.

We have recently seen a step change in the volume of data, the range of data formats and access modes plus a huge increase in the speed at which many data sources are updated. Consequently, information systems can no longer be represented by monolithic, rigorously defined and centralised data models. In many cases, this means that the data relevant to a decision is scattered over multiple, poorly structured, locations such as personal file stores, networked databases, various web pages, etc. Human intelligence is relatively good at working with incomplete knowledge - memorably summarised as the "known knowns, the known unknowns and the unknown unknowns". This talk will look at the strengths and weaknesses of Computational Intelligence in dealing with the problems of the "unknown" and its use in developing tools that can support today’s knowledge workers.

Dr. Pamela Abbott is a lecturer and researcher in the area of Information Systems Management within the Department of Information Systems and Computing, Brunel University, West London.

In achieving success in global sourcing arrangements, the role of a cultural liaison, boundary spanner or transnational intermediary is frequently highlighted as being critical. In this paper, we argue that concepts like "boundary spanning" have been limited in theorizing the complexities of cross-cultural collaborations in offshore outsourcing processes. This paper presents an alternative framework of "creolization" that combines and further extends theoretical understandings of these processes. We investigated 13 companies through 26 in-depth, semi-structured interviews in Xi'an Software Park, an emerging Chinese software and services outsourcing hub. A grounded analysis of the data revealed four conceptual groupings for the practices undertaken at these companies, labeled as boundary spanning, mixed identity, network expansion and cultural hybridity. We posit that the process of creolization supports these practices and furthermore provides a unique basis for strategies positioning cross-cultural work from a supplier's perspective.

Mark Girolami holds a Chair in Statistics at University College London (UCL), Department of Statistical Science. He is also an honorary professor in the Department of Computer Science at UCL, is Director of the Centre for Computational Statistics and Machine Learning at UCL, a Fellow of the Institute of Engineering & Technology, and an EPSRC Advanced Research Fellow.

The talk proposes Metropolis adjusted Langevin and Hamiltonian Monte Carlo sampling methods defined on the Riemann manifold to resolve the shortcomings of existing Monte Carlo algorithms when sampling from target densities that may be high dimensional and exhibit strong correlations. The methods provide fully automated adaptation mechanisms that circumvent the costly pilot runs that are required to tune proposal densities for Metropolis-Hastings or indeed Hamiltonian Monte Carlo and Metropolis adjusted Langevin algorithms. This allows for highly efficient sampling even in very high dimensions where different scalings may be required for the transient and stationary phases of the Markov chain. The methodology proposed exploits the Riemann geometry of the parameter space of statistical models and thus automatically adapts to the local structure when simulating paths across this manifold, providing highly efficient convergence and exploration of the target density. The performance of these Riemann manifold Monte Carlo methods is rigorously assessed by performing inference on logistic regression models, mixture models, log-Gaussian Cox point processes, stochastic volatility models and Bayesian estimation of dynamic systems described by non-linear differential equations. Substantial improvements in the time-normalized effective sample size are reported when compared with alternative sampling approaches.

Dr. Stephanie Schwan is a researcher at the Process Technology & Engineering department, Evonik Industries, Hanau, Germany. Since 2000 she has been working in the area of Quality Engineering. Currently she is at the secondment at Bournemouth University within the framework of INFER (computational Intelligence platform For Evolving and Robust predictive Systems) project.

Stephanie will talk about the challenges in the development and maintenance of soft sensors at Evonik Industries. The research conducted during Stephanie’s three month long secondment will focuse on the identification, development and research of existing and new pre-processing and predictive techniques driven specifically by the process industry requirements though applicable in the general platform context. She will present the topics she wants to focus on as well as the expected outcomes of her secondment.

Rapid development in computer and sensor technology not only used for highly specialised applications but widespread and pervasive across a wide range of business and industry has facilitated easy capture and storage of immense amounts of data. Examples of such data collection include medical history data in health care, financial data in banking, point of sale data in retail, plant monitoring data based on instant availability of various sensor readings in various industries, or airborne hyperspectral imaging data in natural resources identification to mention only a few. However, with an increasing computer power available at affordable prices and the availability of vast amount of data there is an increasing need for robust methods and systems, which can take advantage of all available information.

In essence there is a need for intelligent and smart adaptive methods but do they really exist? Are there any existing intelligent techniques which are more suitable for certain type of problems than others? How do we select those methods and can we be sure that the method of choice is the best for solving our problem? Do we need a combination of methods and if so then how to best combine them for different purposes? Are there any generic frameworks and requirements which would be highly desirable for solving data intensive and unstationary problems? All these questions and many others have been the focus of research vigorously pursued in many disciplines and some of them will be discussed in the talk and have been addressed in greater detail in our recently compiled book with the same title: "Do Smart Adaptive Systems Exist?".

One of the more promising approaches to constructing smart adaptive systems is based on intelligent technologies including artificial neural networks, fuzzy systems, methods from machine learning, parts of learning theory and evolutionary computing which have been especially successful in applications where input-output data can be collected but the underlying physical model is unknown. The incorporation of intelligent technologies has been used in the conception and design of complex systems in which analytical and expert systems techniques are used in combination. Viewed from a much broader perspective, the above mentioned intelligent technologies are constituents of a very active research area known under the names of soft computing, computational intelligence or hybrid intelligent systems.

However, hybrid soft computing frameworks are relatively young, even comparing to the individual constituent technologies, and a lot of research is required to understand their strengths and weaknesses. Nevertheless hybridization and combination of intelligent technologies within a flexible open framework seem to be the most promising direction in achieving the truly smart and adaptive systems today.

Despite all the challenges it is unquestionable that smart adaptive intelligent systems and intelligent technology have started to have a huge impact on our everyday life and many applications can already be found in various commercially available products as illustrated in the recent report compiled by one of the world's leading think tank advanced technology organisations and very suggestively titled: "Get smart: How intelligent technology will enhance our world".

All of the above will be covered in this talk and illustrations provided on the basis of on going collaborative research projects and successful completed applications of various intelligent technologies and highly flexible predictive systems in telecommunication, process and airline industries with such large companies as British Telecommunication plc, Lufthansa Systems GmbH and Evonik-Degussa GmbH.

"Selling the right product to the right customer at the right time for the right price to maximise profits" - this is what revenue management is all about. It has become a mainstream business practice with applications in many key industries, like hospitality and transportation industry.

Airline carriers were the first companies to apply revenue management in the 1970s and remain to be one of the most active users until today. This talk will introduce revenue management in general, describing its components and implementation using different examples and scenarios before looking at airline industry in particular. A special focus will be put on forecasting as a critical factor for the success of a revenue management system.

This talk deals with the current development and challenges on the field of soft sensing. Soft sensors are predictive models applied in the process industry. Here, the models can either replace traditional hardware sensors or provide additional information about critical process characteristics such as the quality of the process product. First soft sensors appeared more than two decades ago and since then there was a steady development in this area. Despite this fact, there are still many challenges that prohibited a real break through, which would lead to a widespread application of soft sensors. The reasons for this fact and possible solutions will be discussed during this talk.

Soft sensors are widely used to estimate process variables that are difficult to measure online. However, the predictive accuracy gradually decreases with changes in the state of chemical plants. Regression models can be updated, but if the model is updated with abnormal data, the predictive ability deteriorates. Therefore, we have proposed a new fault detection and classification method using independent component analysis (ICA) and support vector machine (SVM). This method, named ICA-SVM, was applied to the soft sensor in order to increase fault detection ability and predictive accuracy. We could comprehend the state of a plant by using the ICA-SVM model and estimate the objective variable by the regression model, updating it appropriately. In addition, we have proposed a method to estimate the relationships between applicability domains and the accuracy of prediction of soft sensor models quantitatively. The larger the distances to models(DMs), the lower the estimated accuracy of prediction. Hence, the model between DMs and accuracy can separate variations in process variables and y analyzer fault. The proposed methods were applied to industrial plant data and were found to exhibit higher predictive performance and fault detection ability than traditional methods.

The drug-development process is both time-consuming and expensive: it takes an average of 15 years and $800 million to bring a drug to the market. The process of discovering a new drug for a particular disease usually involves High-Throughput Screening (HTS), a mixture of robotics, control software, liquid-handlers and optical readers - it is an expensive and specialist process. Virtual screening is the computational or in silico screening of chemical compounds and complements the HTS process. It can utilise several computational techniques depending on the amount and type of information available about the compounds and the disease target. This talk will give an overview of the computational techniques that can be used to aid the drug-development process and describes the differing methods used for molecular structure data representation.

It is today a traditional exercise to view the end purpose of computer graphics techniques as photorealism, which can be defined as the generation of synthetic images that cannot be distinguished from reality. After decades of research striving for this, and given appropriate resources in hardware, modern renderers can now produce results very close to photographic images. Improved efficiency for this, as well as further advances, is still possible but at the same time there is an increasing amount of research focusing not on approximation of the real world but on the eventual purpose of the depiction and also all of the communicative aspects this can convey, thus influencing a variety of important factors. These can vary from low-level perceptual processes and emotional responses to cognitive workloads and information interpretation. This talk will focus on the evaluation of non-photorealistic rendering styles in the application area of mobile pedestrian navigation, with particular attention to usability. The methodology includes self-reported measures and task-based experiments. Additionally, there will be a brief discussion in the potential use of novel modalities that could be employed for future work in the area such as eye-tracking and BCIs (brainwave computer interfaces).

: The aim of the lecture is to present the interplay between complex network theory and the modern technology-based networks. The Internet, Semantic Web, service networks will be discussed in this context, then synergies between them and biologic, evolutionary and other emergentr natural networks will be shown. The networks will be also discussed as a result of collective actions of independent subjects which form the network to achieve their targets. The models of agent networks, cell automata networks along with dynamic network phenomena (synchronization, reaching consensus, opinion formation will be presented. In the end - the consequences for the future development of technology-based networks will be discussed.

Mashups are a relatively new approach, to combine data from different sources to create valuable information, principally for data aggregation applications. This utilises the potential of the internet and related technologies, to allow users to process tasks collaboratively, and form communities among those with similar interests. We present currently available mashup platforms and key issues to extend data-orentied mashups into process-oriented mashups.

Estimation of the generalisation ability of a classification or regression model is an important issue in the machine learning world, as it indicates expected performance on previously unseen data and is also used for model selection. Currently used generalisation error estimation procedures like cross--validation (CV) or bootstrap are stochastic and thus require multiple repetitions in order to produce reliable results, which is computationally expensive if not prohibitive. The discussed correntropy-based Density Preserving Sampling procedure (DPS) eliminates the need for repeating the error estimation procedure by dividing the available data into subsets, which are guaranteed to be representative of the input dataset. This allows to produce low variance error estimates with accuracy comparable to 10 times repeated cross-validation at a fraction of computations required by CV, which has been investigated using a set of publicly available benchmark datasets and standard classifiers.