Computer-aided diagnosis of health problems and pathological conditions has become a substantial part of medical, biomedical, and computer science research. This paper focuses on the diagnosis of early and progressive dementia, building on the potential of deep learning (DL) models. The proposed computational framework exploits a magnetic resonance imaging (MRI) brain asymmetry biomarker, which has been associated with early dementia, and employs DL architectures for MRI image classification. Identification of early dementia is accomplished by an eight-layered convolutional neural network (CNN) as well as transfer learning of pretrained CNNs from ImageNet. Different instantiations of the proposed CNN architecture are tested. These are equipped with Softmax, support vector machine (SVM), linear discriminant (LD), or [Formula: see text] -nearest neighbor (KNN) classification layers, assembled as a separate classification module, which are attached to the core CNN architecture. The initial imaging data were obtained from the MRI directory of the Alzheimer's disease neuroimaging initiative 3 (ADNI3) database. The independent testing dataset was created using image preprocessing and segmentation algorithms applied to unseen patients' imaging data. The proposed approach demonstrates a 90.12% accuracy in distinguishing patients who are cognitively normal subjects from those who have Alzheimer's disease (AD), and an 86.40% accuracy in detecting early mild cognitive impairment (EMCI).

Children's acquisition of the English past tense has been widely studied as a testing ground for theories of language development, mostly because it comprises a set of quasi-regular mappings. English verbs are of two types: regular verbs, which form their past tense based on a productive rule, and irregular verbs, which form their past tenses through exceptions to that rule. Although many connectionist models exist for capturing language development, few consider individual differences. In this article, we explore the use of populations of artificial neural networks (ANNs) that evolve according to behavioral genetics principles in order to create computational models capable of capturing the population variability exhibited by children in acquiring English past tense verbs. Literature in the field of behavioral genetics views variability in children's learning in terms of genetic and environmental influences. In our model, the effects of genetic influences are simulated through variations in parameters controlling computational properties of ANNs, and the effects of environmental influences are simulated via a filter applied to the training set. This filter alters the quality of information available to the artificial learning system and creates a unique subsample of the training set for each simulated individual. Our approach uses a population of twins to disentangle genetic and environmental influences on past tense performance and to capture the wide range of variability exhibited by children as they learn English past tenses. We use a novel technique to create the population of ANN twins based on the biological processes of meiosis and fertilization. This approach allows modeling of both individual differences and development (within the lifespan of an individual) in a single framework. Finally, our approach permits the application of selection on developmental performance on the quasi-regular task across generations. Setting individual differences within an evolutionary framework is an important and novel contribution of our work. We present an experimental evaluation of this model, focusing on individual differences in performance. The experiments led to several novel findings, including: divergence of population attributes during selection to favor regular verbs, irregular verbs, or both; evidence of canalization, analogous to Waddington's developmental epigenetic landscape, once selection starts targeting a particular aspect of the task domain; and the limiting effect on the power of selection in the face of stochastic selection (roulette wheel), sexual reproduction, and a variable learning environment for each individual. Most notably, the heritability of traits showed an inverse relationship to optimization. Selected traits show lower heritability as the genetic variation of the population reduces. The simulations demonstrate the viability of linking concepts such as heritability of individual differences, cognitive development, and selection over generations within a single computational framework.

Despite the popularity of MOOCs in providing opportunities for socialization, collaboration, and professional improvement, there has been little research exploring them in the context of Software Engineering Education (SEE). The purpose of this study is to provide a better understanding of practices and challenges when developing academic software engineering MOOCs. To this end, we research (i) how MOOCs in SEE are designed and (ii) what learning design considerations educators and learning designers have when developing SEE-related MOOCs. To address the first research question, a systematic mapping is performed to outline the current landscape. The second research question is answered by analyzing and reflecting on practical experiences acquired during the design of a MOOC on Agile Software Development. The design of the course was based on the Pedagogical Design Pattern Framework for MOOCs, an approach to design for learning that was defined in our previous studies. For this course, two environments were considered: (a) course-time on the Tim Tec MOOC platform for deeper engagement with active learning activities, such as Project-based learning; (b) a Facebook group as a social and collaborative learning space, including scenarios based on problem-based learning to activate prior knowledge. Some of the pedagogical strategies adopted to motivate self-regulated learning included self-introduction video, diary, diagnostic self-assessments, development of small projects alone and in pairs. Our findings provide evidence that there are still several technological and pedagogical challenges that need to be addressed so as to enhance the MOOCs learning experience for SEE. Technological issues identified include the development of tools and educational games, which should be integrated into MOOCs, and the use of learning analytics to support motivation, user experience, and more active learning in specific topics, such as Project Management, Agile Methods, and Requirements Engineering. In turn, the most significant and needed improvement to the pedagogical aspects is a re-thinking of the virtual moment in the MOOC platform so as to optimize it through activities that promote active learning and contextualized assessments.

The cloudUPDRS app is a Class I medical device, namely an active transient non-invasive instrument, certified by the Medicines and Healthcare products Regulatory Agency in the UK for the clinical assessment of the motor symptoms of Parkinson's Disease. The app follows closely the Unified Parkinson's Disease Rating Scale which is the most commonly used protocol in the clinical study of PD; can be used by patients and their carers at home or in the community; and, requires the user to perform a sequence of iterated movements which are recorded by the phone sensors. This paper discusses how the cloudUPDRS system addresses two key challenges towards meeting essential consistency and efficiency requirements, namely: (i) How to ensure high-quality data collection especially considering the unsupervised nature of the test, in particular, how to achieve firm user adherence to the prescribed movements; and (ii) How to reduce test duration from approximately 25 minutes typically required by an experienced patient, to below 4 minutes, a threshold identified as critical to obtain significant improvements in clinical compliance. To address the former, we combine a bespoke design of the user experience tailored so as to constrain context, with a deep learning approach used to identify failures to follow the movement protocol while at the same time limiting false positives to avoid unnecessary repetition. We address the latter by developing a machine learning approach to personalise assessments by selecting those elements of the UPDRS protocol that most closely match individual symptom profiles and thus offer the highest inferential power hence closely estimating the patent's overall UPRDS score.

We analyse engagement and performance data arising from participants' interactions with an in-house LMS at Imperial College London while a cohort of students follow two courses on a new online postgraduate degree in Management. We identify and investigate two main questions relating to the relationships between engagement and performance, drawing recommendations for improved guidelines to inform the design of such courses.

Training a multilayer perceptron (MLP) with algorithms employing global search strategies has been an important research direction in the field of neural networks. Despite a number of significant results, an important matter concerning the bounds of the search region---typically defined as a box---where a global optimization method has to search for a potential global minimizer seems to be unresolved. The approach presented in this paper builds on interval analysis and attempts to define guaranteed bounds in the search space prior to applying a global search algorithm for training an MLP. These bounds depend on the machine precision and the term guaranteed denotes that the region defined surely encloses weight sets that are global minimizers of the neural network's error function. Although the solution set to the bounding problem for an MLP is in general non-convex, the paper presents the theoretical results that help deriving a box which is a convex set. This box is an outer approximation of the algebraic solutions to the interval equations resulting from the function implemented by the network nodes. An experimental study using well known benchmarks is presented in accordance with the theoretical results.

Reliable estimation of a neural network's domain of validity is important for a number of reasons such as assessing its ability to cope with a given problem, evaluating the consistency of its generalization etc. In this paper we introduce a new approach to estimate the domain of validity of a neural network based on Set Inversion Via Interval Analysis (SIVIA), the methodology established by Jaulin and Walter [1]. This approach was originally introduced in order to solve nonlinear parameter estimation problems in a bounded error context and proved to be effective in tackling several types of problems dealing with nonlinear systems analysis. The dependence of a neural network output on the pattern data is a nonlinear function and hence derivation of the impact of the input data to the neural network function can be addressed as a nonlinear parameter estimation problem that can be tackled by SIVIA. We present concrete application examples and show how the proposed method allows to delimit the domain of validity of a trained neural network. We discuss advantages, pitfalls and potential improvements offered to neural networks.

In exploratory learning environments, learners can use different strategies to solve the same problem. Not all these strategies, however, are known to the teacher and, even if they were, they need considerable time and effort to introduce them in the knowledge base. In this paper we propose a learning mechanism that extracts strategies from user data and presents them to the teacher for further authoring. To this end, a clustering approach is used in which the strategies of learners are grouped into clusters and the teacher is presented with a representative strategy for each cluster. The teacher can then decide whether to store the proposed strategies or to author them further. This approach allows populating the knowledge base using user data, thus saving authoring time for the teacher.

If a population of programs evolved not for a few hundred generations but for a few hundred thousand or more, could it generate more interesting behaviours and tackle more complex problems? We begin to investigate this question by introducing Tweaking Mutation Behaviour Learning (TMBL), a form of evolutionary computation designed to meet this challenge. Whereas Genetic Programming (GP) typically involves creating a large pool of initial solutions and then shuffling them (with crossover and mutation) over relatively few generations, TMBL focuses on the cumulative acquisition of small adaptive mutations over many generations. In particular, we aim to reduce limits on long term fitness growth by encouraging tweaks: changes which affect behaviour without ruining the existing functionality. We use this notion to construct a standard representation for TMBL. We then experimentally compare TMBL against linear GP and tree-based GP and find that TMBL shows strong signs of being more conducive to the long term growth of fitness.

Semantic web technologies and autonomic computing principles are combined in this paper in an attempt to design and build a learning design environment that possesses context-aware features. Our approach builds on the features of self-management and organisation of autonomic computing but uses self-configuration as a means to extend a knowledge-based inference through the design of meta-level inference. Thus, the context inference is modelled using a meta-interpreter and self-configuration rules. The details of our approach are presented demonstrating the use of self-configurable inferencing to support the creation and use of context-paths across learning design domain concepts. The paths exploit ontology alignment principles to determine contextual relevance between user learning designs and core system knowledge.

The nature of the activities that take place in Exploratory Learning Environments allow generating a variety of learner trajectories and makes difficult to develop a model of all possible behaviours (correct or incorrect). To alleviate this situation, we propose an approach for knowledge representation and identification of strategies followed by the learners during exploratory learning. Our approach combines heterogeneous sources of information and defines appropriate similarity measures for strategy identification. Some scenarios from the domain of mathematical generalisation are presented.

Book synopsis: This publication covers papers presented at the Artificial Intelligence in Education conference 2009 (AIED). AIED2009 is part of an ongoing series of biennial international conferences for top quality research in intelligent systems and cognitive science for educational computing applications. The conference provides opportunities for the cross-fertilization of techniques from many fields that make up this interdisciplinary research area, including: artificial intelligence, computer science, cognitive and learning sciences, education, educational technology, psychology, philosophy, sociology, anthropology, linguistics and the many domain-specific areas for which AIED systems have been designed and evaluated. AIED2009 focuses on the theme 'Building learning systems that care: from knowledge representation to affective modelling'. The key research question is how to tackle the complex issues related to building learning systems that care, ranging from representing knowledge and context to modelling social, cognitive, metacognitive and affective dimensions.This requires multidisciplinary research that links theory and technology from artificial intelligence, cognitive science and computer science with theory and practice from education and the social sciences.

In this paper, we describe our work to investigate how much cyclic graph based Genetic Programming (GP) can be accelerated on one machine using currently available mid-range Graphics Processing Units (GPUs).

Several forms of computer program (or representation) have been proposed for Genetic Programming (GP) systems to evolve, such as linear, tree based or graph based. Typically, GP representations are highly effective during the initial search phases of evolution but stagnate before deep levels of complexity are acquired. A new representation, TREAD, is proposed to combine aspects of flow of execution and flow of data systems. The distinguishing features of TREAD are designed for researching improvements to the long term acquisition of novel features in GP (at the expense of the speed of the initial search if necessary). TREAD is validated on a symbolic regression problem and is found to be capable of successfully developing solutions through artificial evolution.

No abstract available.

Recurrent networks constitute an elegant way of increasing the capacity of feedforward networks to deal with complex data in the form of sequences of vectors. They are well known for their power to model temporal dependencies and process sequences for classification, recognition, and transduction. In this paper, we propose a nonmonotone conjugate gradient training algorithm for recurrent neural networks, which is equipped with an adaptive tuning strategy for the nonmonotone learning horizon. Simulation results show that this modification of conjugate gradient is more effective than the original CG in four applications using three different recurrent network architectures.

The rapid growth of the web, wireless communication, hypermedia, and the data interoperability among massive digital repositories will enable virtually everyone to have access to huge amounts of content on almost any topic, and services. In this context, personalised systems have emerged as environments that adapt content, structure, and/or presentation to individual user characteristics, usage behaviour and user needs. Adaptation is defined as the concept of making adjustments in a system to accommodate diversity in user requirements and abilities in order to maintain the appropriate context for interaction. It is acknowledged that individuals differ in traits, such as skills, aptitudes and preferences for processing information, constructing knowledge from information, and applying it to real-world problem solving. Personalised systems aim to enhance the mode of delivery of information to the user, increase both flexibility and adaptation of the e-content and of the interface to individual needs, requirements and preferences. Personalisation technologies enable organisations to develop systems, offer added value services and facilitate resource access in several application areas. This direction in web-based systems development has created various challenges and opportunities in terms of adaptation techniques, customisation and personalisation of content, navigation support and layout. New architectures for personalised systems and adaptation techniques have been developed in order to take full advantage of the potential of this new approach to tailor content, navigation support and services to user needs. This kind of technology can be applied on the current infrastructure of the web, or exploit the Internet, to personalise learning environments, recommender systems, and services to customers. It may also be facilitated by recent advances in the area of semantic web.

Abstract Search engines are very popular tools for collecting information from distributed resources. They provide not only search facilities, but they also offer directories for users to browse content divided into groups. In this paper, we've adopted an individual differences approach to explore user's attitudes towards various interface features provided by existing Web directories. Among a variety of individual differences, cognitive style is a particularly important characteristic that influences the effectiveness of information seeking. Empirical results indicate that users' cognitive styles influence their reactions to the organization of subject categories, presentation of the results, and screen layout. We developed a set of design guidelines on the basis of these results, and propose a flexible interface that adopts these guidelines to accommodate the preferences of different cognitive style groups.

We address the problem of integrating user preferences with network quality of service parameters for the streaming of media content, and suggest protocol stack configurations that satisfy user and technical requirements to the best available degree. Our approach is able to handle inconsistencies between user and networking considerations, formulating the problem of construction of tailor-made protocols as a prioritization problem, solvable using fuzzy programming.

This paper proposes an approach that integrates technical concerns with user perceptual considerations for intelligent decision making in the construction of tailor-made multimedia communication protocols. Thus, the proposed approach, based on multicriteria decision making (MDM), incorporates not only classical networking considerations, but, indeed, user preferences as well. Furthermore, in keeping with the task-dependent nature consistently identified in multimedia scenarios, the suggested communication protocols also take into account the type of multimedia application that they are transporting. Lastly, this approach also opens the possibility for such protocols to dynamically adapt based on a changing operating environment and user's preferences.

In this paper the fuzzy knowledge representation of a neural network-based fuzzy model is presented. The model is used to assess student's motivational state in a discovery learning environment. Student's observable behavior and motivational factors are described with linguistic variables. The inputs of the model are tailored from real students' data, with the assistance of a group of expert teachers. Results of our preliminary study were encouraging, since data obtained from real students' log files, have been successfully used to form the membership functions that assign membership degrees to the linguistic values of the linguistic variables.

A challenging situation gradient-based learning algorithms encounter is an occasional converge to undesired local minima. To alleviate this situation, this paper builds on the theory of nonextensive statistical mechanics to develop a new adaptive gradient-based learning scheme that applies a sign-based weight adjustment, inspired from the Rprop algorithm, on a perturbed version of the original error function. The perturbations are characterized by the q entropic index of the nonextensive entropy, and their impact is controlled by means of regularization. This approach modifies the error landscape at each iteration allowing the algorithm to explore previously unavailable regions of the error surface, and possibly escape undesired local minima. The performance of the adaptive scheme is empirically evaluated using problems from the UCI repository of machine learning databases and other classic benchmarks.

A novel generalized theoretical result is presented that underpins the development of globally convergent first-order batch training algorithms which employ local learning rates. This result allows us to equip algorithms of this class with a strategy for adapting the overall direction of search to a descent one. In this way, a decrease of the batch-error measure at each training iteration is ensured, and convergence of the sequence of weight iterates to a local minimizer of the batch error function is obtained from remote initial weights. The effectiveness of the theoretical result is illustrated in three application examples by comparing two well-known training algorithms with local learning rates to their globally convergent modifications.

We present deterministic nonmonotone learning strategies for multilayer perceptrons (MLPs), i.e., deterministic training algorithms in which error function values are allowed to increase at some epochs. To this end, we argue that the current error function value must satisfy a nonmonotone criterion with respect to the maximum error function value of the M previous epochs, and we propose a subprocedure to dynamically compute M. The nonmonotone strategy can be incorporated in any batch training algorithm and provides fast, stable, and reliable learning. Experimental results in different classes of problems show that this approach improves the convergence speed and success percentage of first-order training algorithms and alleviates the need for fine-tuning problem-depended heuristic parameters.

A few approaches have been presented in the literature towards the discrimination of texture in medical images. Medical experts proposed that the more valuable information for discriminating among normal and suspicious cancer regions in endoscopic images is the texture of the examined tissue. Texture can be encoded by a number of mathematical descriptors. Three well-known textural descriptors, as well as a new wavelet-based one are used in this paper for an accurate study and evaluation of the methodologies encountered. Experiments conducted include tests with various images from the Brodatz album, as well as interpretation of tissue regions in endoscopic image. In all cases the recognition task is supported by multilayer perceptron type neural network architectures.

The vision of a new generation of network communication architectures, which deliver a Quality of Service based on intelligent decisions about the interactions that typically take place in a multimedia scenario, encourages researchers to look at novel ways of matching user-level requirements with parameters characterising underlying network performance. In this paper, we suggest an integrated architecture that makes use of the objective-technical information provided by the designer and the subjective-perceptual information supplied by the user for intelligent decision making in the construction of communication protocols. Thus, this approach, based on the Analytic Hierarchy Process, incorporates not only classical Quality of Service considerations, but, indeed, user preferences as well. Furthermore, in keeping with the task-dependent nature consistently identified in multimedia scenarios, the suggested communication protocols also take into account the type of multimedia application, which they are transporting. Lastly, our approach also opens the possibility for such protocols to dynamically adapt based on a changing operating environment. 1.

The concept of quality of service (QoS) in distributed multimedia systems is indelibly associated with the provision of an acceptable level of application performance. In turn this performance is itself dependent on both the perceived, subjective quality of the application as well as its robustness to network congestion. We suggest a framework for QoS management that incorporates not only classical QoS considerations, but, user preferences as well. Our approach constructs an appropriately tailored communication protocol using objective-technical judgements provided by the designer and subjective-perceptual judgements supplied by the user for intelligent decision making. Due to the nature of our problem, where these two information facets may be inconsistent, the need to apply a weight determination technique suitable to handle inconsistencies was indispensable. To this end, a fuzzy logic-based description of the problem has been developed and fuzzy programming was used to obtain a user-oriented ordering of QoS parameters.

"Model-based design and evaluation of interactive applications." European Journal of Information Systems, 10(3), p. 175

A new theorem for the development and convergence analysis of supervised training algorithms with an adaptive learning rate for each weight is presented. Based on this theoretical result, a strategy is proposed to automatically adapt the search direction, as well as the step-size length along the resultant search direction. This strategy is applied to some well known local learning algorithms to investigate its effectiveness.

In this paper neuro-fuzzy synergism is applied to implement content sequencing in adaptive hypermedia systems. The level of understanding of the learner is used to construct lessons adapted to the learner's knowledge goals and level of expertise on the domain concepts s/he has already studied. The learner's evaluation is based on defining appropriate fuzzy sets and relate learner's response with appropriate knowledge and cognitive characterizations. A connectionist-based structure of the domain knowledge is adopted for representing knowledge and inferring the planning strategy for generating the hypermedia page from pieces of educational material. The fuzziness associated with the evaluation of the learner is handled well by the proposed connectionist architecture.

A neuro-fuzzy approach is introduced to implement lesson adaptation in a Web-based course. Several key points that affect the effectiveness of an adaptive learning environment are investigated the development of the educational material, the structure of the domain knowledge, the instructional design and the evaluation of the learner knowledge under uncertainty. The proposed approach allows the generation of the content of a hypermedia page from pieces of educational material based on goal-oriented teaching and making use of the background knowledge of the learner.

Training algorithms suitable to work under imprecise conditions are proposed. They require only the algebraic sign of the error function or its gradient to be correct, and depending on the way they update the weights, they are analyzed as composite nonlinear successive overrelaxation (SOR) methods or composite nonlinear Jacobi methods, applied to the gradient of the error function. The local convergence behavior of the proposed algorithms is also studied. The proposed approach seems practically useful when training is affected by technology imperfections, limited precision in operations and data, hardware component variations and environmental changes that cause unpredictable deviations of parameter values from the designed configuration. Therefore, it may be difficult or impossible to obtain very precise values for the error function and the gradient of the error during training.

We present nonmonotone methods for feedforward neural network training, i.e., training methods in which error function values are allowed to increase at some iterations. More specifically, at each epoch we impose that the current error function value must satisfy an Armijo-type criterion, with respect to the maximum error function value of M previous epochs. A strategy to dynamically adapt M is suggested and two training algorithms with adaptive learning rates that successfully employ the above mentioned acceptability criterion are proposed. Experimental results show that the nonmonotone learning strategy improves the convergence speed and the success rate of the methods considered.

The issue of implementing adaptation in a Web-based course using an adaptive lesson presentation (or content-level adaptation) method is a promising direction of research. The goal is to adapt dynamically the supplied educational material so that it meets the individual educational needs of each particular learner. An important aspect in producing a learner-adapted system is the structuring of the domain knowledge in such a way that it will be possible to do adaptations. In this paper, a connectionist-based structure of the domain knowledge model is adopted. This approach allows generating the content of a hypermedia page from pieces of educational material based on a goal-oriented way of teaching and making use of the background knowledge of the learner.

An empirical approach that makes use of neuro-fuzzy synergism to evaluate the students in the context of an intelligent tutoring system is presented. In this way, a qualitative model of the student is generated, which is able to evaluate information regarding student's knowledge and cognitive abilities in a domain area. The neuro-fuzzy model has been tested on a prototype tutoring system in the physics domain of the vertical projectory motions and the results have been very satisfactory.

A mathematical framework for the convergence analysis of the well known Quickprop method is described. The convergence of this method is analyzed. Furthermore, we present modifications of the algorithm that exhibit improved convergence speed and stability and at the same time, alleviate the use of heuristic learning parameters. Simulations are conducted to compare and evaluate the performance of a proposed modified Quickprop algorithm with various popular training algorithms. The results of the experiments indicate that the increased convergence rates, achieved by the proposed algorithm, affect by no means its generalization capability and stability.

This article focuses on gradient-based backpropagation algorithms that use either a common adaptive learning rate for all weights or an individual adaptive learning rate for each weight and apply the Goldstein/Armijo line search. The learning-rate adaptation is based on descent techniques and estimates of the local Lipschitz constant that are obtained without additional error function and gradient evaluations. The proposed algorithms improve the backpropagation training in terms of both convergence rate and convergence characteristics, such as stable learning and robustness to oscillations. Simulations are conducted to compare and evaluate the convergence behavior of these gradient-based training algorithms with several popular training methods.

A hybrid model for representing knowledge in a Greek natural language interface to MS-DOS is presented. This model employs semantic knowledge of the operating system domain to develop a hybrid architecture integrating symbolic and connectionist modules for recognizing the MS-DOS command labels of related or unrelated successive natural language user messages. This approach was deeply motivated by handling ellipsis, ambiguity and flexibility of sentence structure as well as possible. A comparative experimental evaluation of this approach to a symbolic one illustrating their pros and cons is also given. Finally, ways of extending this model are proposed with the view of making the natural language interface more user friendly and functional.