Encouraging results of last years in the field of knowledge representation within virtual learning environments confirms that artificial intelligence research in this topic find it very beneficial to integrate the knowledge psychological research have accumulated on understanding the cognitive mechanism of human learning and all the positive results obtained in computational modelling theories. This chapter introduces a novel cognitive and computational knowledge representation approach inspired by cognitive theories which explain the human cognitive activity in terms of memory subsystems and their processes, and whose aim is to suggest formal computational models of knowledge that offer efficient and expressive representation structures for virtual learning. Practical studies both contribute to validate the novel approach and permit to draw general conclusions.

In wireless communication networks, it is desirable to achieve energy efficiency, while maintaining quality of service. Overall, energy efficiency can be achieved by minimizing the power output for each communication device. This paper presents a novel energy-saving adaptive transmit power control (TPC) algorithm based on bit error rate (BER) feedback (BER-TPC) that outperforms conventional approaches based on signal-to-interference-plus-noise ratio (SINR) feedback (SINR-TPC). It is a distributed algorithm deployed in both transmitting and receiving devices that can be applied for various wireless network topologies and protocols. This paper addresses the power efficiency of adaptive TPC, in terms of reduced total transmit power of the network by smoothing transmit power transients during adaptive iterations. It is achieved using a distributed closed-loop power control system that applies heuristically estimated dual-rate power adjustments during adaptive iterations. Because the proposed system uses a mathematical formula for power adjustment, operational bounds for stability can be provided using simple analysis. It is demonstrated that the proposed system provides adequate tradeoff between performance and complexity, in terms of reduction in sensitivities and better tracking performance in following the desired reference power curves used in the simulations. Case scenarios are simulated demonstrating approximately 1.39-dB transmit power savings compared to conventional methods.

Model-tracing tutors (MTTs) have proven effective for the tutoring of well-defined tasks, but the pedagogical interventions they produce are limited and usually require the inclusion of pedagogical content, such as text message templates, in the model of the task. The capability to generate pedagogical content would be beneficial to MTT frameworks, as it would lessen the task-specific efforts and could lead to the capability of providing more sophisticated pedagogical interventions. In this paper, we show how Astus, as an MTT framework, strive to attain a higher level of automation when generating pedagogical interventions compared to other MTT frameworks such as TDK and CTAT's MTTs. This is achieved by designing a knowledge representation approach in which each type of knowledge unit has a clearly defined semantic on which the MTT's pedagogical module can rely on. We explain how this knowledge representation approach is implemented as a knowledge-based system in ASTUS and show how it allows the development of MTTs that can automatically generate the pedagogical content required to provide next-step hints and negative feedback on errors. Multiple small-scale experiments were conducted with computer science undergraduate students in order to obtain a preliminary assessment of the effectiveness of Astus's pedagogical interventions.

In this paper we introduce a novel bit error rate (BER) feedback transmit power control (TPC) system using a first-time third order quadratic approximation of the power–time curve. This approximation improves the power efficiency of a dual-rate TPC algorithm in terms of reduced total transmit power by smoothing transmit power transients during adaptive iterations. We show that the third order approximator outperforms linear and second order approximators in terms of transmit power savings, sensitivities, error magnitude, and better tracking performance in following reference desired power curves. For the approximator, we determine operational bounds for stability, and demonstrate algorithm behavior using critical valued inputs. In addition we demonstrate value in using a dynamic, rather than static, performance benchmark for quality of service approximation (obtained used scaled maximum acceptable BER), and provide heuristic estimates for the input parameters for the dynamic benchmark.

We propose a new efficient algorithm for solving the cluster labeling problem in support vector clustering (SVC). The proposed algorithm analyzes the topology of the function describing the SVC cluster contours and explores interconnection paths between critical points separating distinct cluster contours. This process allows distinguishing disjoint clusters and associating each point to its respective one. The proposed algorithm implements a new fast method for detecting and classifying critical points while analyzing the interconnection patterns between them. Experiments indicate that the proposed algorithm significantly improves the accuracy of the SVC labeling process in the presence of clusters of complex shape, while reducing the processing time required by existing SVC labeling algorithms by orders of magnitude.

Increasingly, unlicensed devices demand access to the licensed wireless spectrum, raising the problem of spectrum scarcity. This problem can be addressed by allowing unlicensed devices opportunistic access. Because opportunistic devices regard licensed users as interference, channel Signal to Interference Noise Ratio (SINR) is an important consideration for transmission. Cognitive Radio (CR) technology allows devices to sense spectrum and dynamically allocate bits to channels with good SINR. In this paper we introduce a stochastic modulation algorithm called the Cognitive Radio Stochastic Modulation (CRSM), for use as an extension to established M-ary Modulation schemes in underlay channels with poor SINR. In this modulation scheme, in additive channels, both the transmitting and receiving CR sense the magnitude and variance of the interference energy in the channel. The transmitter uses deterministic signals as unique identifiers in manipulating the variance of the signal detected at the receiver. The detected change in variance, measured at the receiver, is indicative of the presence or absence of a signal. In this paper we determine the critical regions for the CRSM and introduce a probability of error model.

Personal Area Network (PAN) Devices are often components of communication systems which opportunistically access the wireless spectrum. As such, they must operate without presenting interference to licensed users, while meeting their individual, varied Quality of Service (QoS) requirements. In this paper we use a novel Transmit Power Control (TPC) algorithm to reduce interference. In theory, over the life of a Secondary User's (SU's) transmission, there is an optimal power time curve reflective of the minimum amount of power necessary with which a device can transmit while maintaining acceptable BER, and practicing Interference Avoidance (IA). Using the response of the algorithm to instantaneous channel conditions as well as an iteratively updated QoS benchmark, we obtain a quadratic approximation of the desired power-time curve. Stability of the quadratic approximation is demonstrated through Matlab simulations using critical valued inputs to the approximator. From our results, we also show that our quadratic approximator outperforms others of different orders in terms of reduction in sensitivities to relatively small changes in inputs, and better tracking performance in following reference desired power curves used in the simulations. Our algorithm reduces transmit power (and thus interference) by approximately 3.5 dB compared to conventional methods while maintaining the required QoS.

Problem-solving or step-based ITS have been proven successful for well-defined domains, particularly in well-defined tasks, but their success is mitigated by their cost. Typically, the main factor behind the cost is the efforts needed to model the task domain. Different approaches have been investigated to reduce these efforts: Model-Tracing Tutors (e.g. Cognitive Tutors [1], Andes [2]), Constraint-Based Tutors (e.g. SQL-Tutor [3], ASPIRE [4]) and Example-Tracing Tutors (e.g. CTAT [5], ASSISTment [6]).

To assist learners during problem-solving activities, an intelligent tutoring system (ITS) has to be equipped with domain knowledge that can support appropriate tutoring services. Providing domain knowledge is usually done by adopting one of the following paradigms: building a cognitive model, specifying constraints, integrating an expert system, and using data mining algorithms to learn domain knowledge. However, for some ill-defined domains, each single paradigm may present some advantages and limitations in terms of the required resources for deploying it, and tutoring support that can be offered. To address this issue, we propose using a multiparadigm approach. In this paper, we explain how we have applied this idea in CanadarmTutor, an ITS for learning to operate the Canadarm2 robotic arm. To support tutoring services in this ill-defined domain, we have developed a multiparadigm model combining: 1) a cognitive model to cover well-defined parts of the task and spatial reasoning, 2) a data mining approach for automatically building a task model from user solutions for ill-defined parts of the task, and 3) a 3D path-planner to cover other parts of the task for which no user data are available. The multiparadigm version of CanadarmTutor allows providing a richer set of tutoring services than what could be offered with previous single paradigm versions of CanadarmTutor.

Model-tracing tutors were shown to be effective for the tutoring of problem solving tasks, but they usually lack the capability to provide feedback on learners’ off-path steps. In this paper, we define a method, inspired by Sierra, to diagnose many of the learners’ errors from their off-path steps. This method is implemented in Astus, a model-tracing tutor authoring framework. We show how Astus diagnose errors from off-path steps and use the resulting diagnostic to generate negative feedback.

ASTUS is an authoring framework designed to create model-tracing tutors with similar efforts to those needed to create Cognitive Tutors. Its knowledge representation system was designed to model the teacher's point of view of the task and to be manipulated by task independent processes such as the automatic generation of sophisticated pedagogical feedback. The first type of feedback we automated is instructions provided as next step hints. Whereas next step hints are classically authored by teachers and integrated in the model of the task, our framework automatically generates them from task independent templates. In this paper, we explain, using examples taken from a floating-point number conversion tutor, how our knowledge representation approach facilitates the generation of next-step hints. We then present experiments, conducted to validate our approach, showing that generated hints can be as efficient and appreciated as teacher authored ones.

Cette dissertation présente un ensemble d'algorithmes visant à en permettre un usage rapide, robuste et automatique des « Support Vector Machines » (SVM) non supervisés dans un contexte d'analyse de données. Les SVM non supervisés se déclinent sous deux types algorithmes prometteurs, le « Support Vector Clustering » (SVC) et le « Support Vector Domain Description » (SVDD), offrant respectivement une solution à deux problèmes importants en analyse de données, soit la recherche de groupements homogènes (« clustering »), ainsi que la reconnaissance d'éléments atypiques (« novelty/abnomaly detection ») à partir d'un ensemble de données. Cette recherche propose des solutions concrètes à trois limitations fondamentales inhérentes à ces deux algorithmes, notamment 1) l'absence d'algorithme d'optimisation efficace permettant d'exécuter la phase d'entrainement des SVDD et SVC sur des ensembles de données volumineux dans un délai acceptable, 2) le manque d'efficacité et de robustesse des algorithmes existants de partitionnement des données pour SVC, ainsi que 3) l'absence de stratégies de sélection automatique des hyperparamètres pour SVDD et SVC contrôlant la complexité et la tolérance au bruit des modèles générés. La résolution individuelle des trois limitations mentionnées précédemment constitue les trois axes principaux de cette thèse doctorale, chacun faisant l'objet d'un article scientifique proposant des stratégies et algorithmes permettant un usage rapide, robuste et exempt de paramètres d'entrée des SVDD et SVC sur des ensembles de données arbitraires.

Modeling learners is a fundamental part of intelligent tutoring systems. It allows tutors to provide personalized feedback and to assess the learners' mastery over a task domain. One aspect often overlooked is the modeling of erroneous behaviors that can be used to provide error specific feedback. This is especially true for model-tracing tutors that usually require erroneous procedural knowledge associated to each of the possible error. This process can be automated thanks to a task independent model describing the learners' erroneous behaviors. The model proposed in this paper is inspired by the Sierra theory of procedural error and is developed for ASTUS, an authoring framework for model-tracing tutors.

A personal bankruptcy prediction system running on credit card data is proposed. Personal bankruptcy, which usually results in significant losses to creditors, is a rapidly increasing yet little understood phenomenon. The most commonly used methods in personal bankruptcy prediction are credit scoring models. Some data mining models have also been investigated in this domain. Neither the scoring models nor the existing data mining methods adequately take sequence information in credit card data into account. In our system, sequence patterns, obtained by developing sequence mining techniques and applying them to credit card data from one major Canadian bank, are employed as main predictors. The mined sequence patterns, which we refer to as bankruptcy features, are represented in low-dimensional vector space. From the new feature space, which can be extended with some existing prediction-capable features (e.g., credit score), a support vector machine (SVM) classifier is built to combine these mined and already existing features. Our system is readily comprehensible and demonstrates promising prediction performance. (c) 2012 Elsevier Ltd. All rights reserved.

Building an intelligent tutoring system requires to define an expertise model that can support appropriate tutoring services. This is usually done by adopting one of the following paradigms: building a cognitive model, specifying constraints, integrating an expert system and using data mining algorithms to learn domain knowledge. However, for some ill-defined domains, the use of a single paradigm could lead to a weak support of the user in terms of tutoring feedback. To address, this issue, we propose to use a multi-paradigm approach. We illustrate this idea in a tutoring system for robotic arm manipulation training. To support tutoring services in this ill-defined domain, we have developed a multi-paradigm model combining: (1) a data mining approach for automatically building a task model from user solutions, (2) a cognitive model to cover well-defined parts of the task and spatial reasoning, (3) and a 3D path-planner to cover all other aspects of the task. Experimental results indicate that the multi-paradigm approach allows providing assistance to learners that is much richer than what is offered with each single paradigm.

Clustering categorical data poses two challenges defining an inherently meaningful similarity measure, and effectively dealing with clusters which are often embedded in different subspaces. In this paper, we propose a novel divisive hierarchical clustering algorithm for categorical data, named DHCC. We view the task of clustering categorical data from an optimization perspective, and propose effective procedures to initialize and refine the splitting of clusters. The initialization of the splitting is based on multiple correspondence analysis (MCA). We also devise a strategy for deciding when to terminate the splitting process. The proposed algorithm has five merits. First, due to its hierarchical nature, our algorithm yields a dendrogram representing nested groupings of patterns and similarity levels at different granularities. Second, it is parameter-free, fully automatic and, in particular, requires no assumption regarding the number of clusters. Third, it is independent of the order in which the data is processed. Fourth, it is scalable to large data sets. And finally, our algorithm is capable of seamlessly discovering clusters embedded in subspaces, thanks to its use of a novel data representation and Chi-square dissimilarity measures. Experiments on both synthetic and real data demonstrate the superior performance of our algorithm.

The chapter presents a novel modular adaptive artful intelligent assistance system for cognitively and/or memory impaired people engaged in the realisation of their activities of daily living (ADLs). The goal of this assistance system is to help disabled persons moving/evolving within a controlled environment in order to provide logistic support in achieving their ADLs. Empirical results of practical tests are presented and interpreted. Some deductions about the key features that represent originalities of the assistance system are drawn and future works are announced.

Step-based ITS have been proven successful for well-defined domains, particularly in well-defined tasks, but their success is mitigated by the amount of effort needed to build them. Typically, the main factor behind these efforts is the model of the task domain. Different approaches have been investigated to reduce these efforts: Model-Tracing Tutors (e.g. Cognitive Tutors, Andes), Constraint-Based Tutors (e.g. SQL-Tutor, ASPIRE) and Example-Tracing Tutors (e.g. CTAT’s, ASSISTment).

The objective of this paper is to study various methods of maintaining user defined Quality of Service (QoS) in a wireless Cognitive Radio (CR) System. The agility of CRs can be used to make adjustments to Bit Error Rate (BER), Bit Rate (BR) and transmit Power Spectral Density (PSD). We present a modified channel capacity analytic formula, methods for a power control algorithm, and methods for decision making on spectrum allocation. We show how parameters such as BER, Signal to Interference Noise Ratio (SINR) thresholds, and analysis and design based on our modified channel capacity formula can be subtly manipulated to affect QoS. The power control algorithm is part of a system that maintains user defined BER and BR in the presence of external user interference. This system is a design tool with which to achieve predictable and consistent QoS.