D.M. Russell

Intelligent and Adaptive Systems

Intelligence and adaptation have been very important factors in designing learning systems. While the research in the field has yielded much useful information, a lot more remains to be done. We suggest that adopting a granular structure reduces some of the complexity and consideration of the teacher as an environmental context of the system is crucial for its success. The discussion is based on our experience of designing and implementing such a system and the feedback from a comparative independent study.

An initial framework of contexts for designing usable intelligent tutoring systems

The notion of context has been an issue of research in various aspects of intelligent systems such as knowledge management, natural language processing, reasoning and so on. This paper focuses on the various contexts surrounding the design and use of Intelligent Tutoring Systems (ITS) and proposes an initial framework of contexts by classifying them into three major groupingsc interactional, environmental and objectival contexts. Interactional contexts are used by the system, environmental contexts surround its design and use while objectival contexts refer to the objectives of an educational system as exhibited by its ‘teaching’ and ‘assessment’ practices. A better understanding of these contexts is essential for designing better and more usable intelligent tutoring systems.

A computer-based intelligent assessment system for numeric disciplines

The paper describes an intelligent assessment system for numeric disciplines. The system works in conjunction with the intelligent tutoring tools developed by TLTP Byzantium, a consortium of six UK Universities. The benefits of the intelligent assessment system discussed in this paper include the saving of teacher time and effort previously spent in marking and compilation of results. The faster turnaround of the assessment related work resulting into a much shorter testing, assessment and feedback cycle, enables more frequent testing. Since tutoring tools are knowledge based, they are capable of generating infinite number of test problems by randomly selecting the independent variables and assigning them random values, as well as providing solution to the generated problems. For developing data interpretation skills, it is possible for a teacher to hand out a problem expressed in a narrative form and provide a model answer to the assessment system. A unique feature of the Byzantium assessment system is its capability of discriminating between incorrect interpretation of given data and incorrect method of solution, allowing a teacher to set a fractional score for a variable that is calculated using a correct method but based on an incorrect interpretation of data.

Predicting corporate bankruptcy using modular neural networks

The paper reports on the use of modular neural networks for predicting corporate bankruptcy. We obtained our financial, as well as, political and economic data from The London Stock Exchange, JORDANS financial database of major British public and private companies, and the Bank of England. In the past, various statistical techniques, such as univariate and multivariate discriminant analysis have been used in the modelling of corporate bankruptcy prediction. We use domain expert knowledge to select, and organise data in the modular neural network architecture constructed for this study. There are three sub-networks representing the periods, 1994, 1995, and 1996. Each sub-network is made of five adjacent networks representing the Balance Sheet network, the Profit and Loss network, the Financial Summary network, the Key Financial Ratios network, and the Economic and Political factors network. These adjacent networks although coupled but not linked at the input level represent five facets of failure in predicting corporate bankruptcy. The training sets represent data for 2500 companies selected randomly from a population of 270000 sample. The trained neural network will access 435000 data records before making a prediction for the particular company. The results obtained shows that neural networks outperform statistical techniques in modelling corporate failure prediction.

Intelligent Tutoring Systems: Confluence of Information Science and Cognitive Science

The advent of Internet as a global communication medium has brought a new focus on an area of research in designing Intelligent Tutoring System (ITS) that has not been adequately considered so far. In the main, this has been due to the localised nature of most academic environments limiting the sources of information and an implicit assumption that information and knowledge are synonymous. These factors have led to overemphasis on learner modelling in the traditional ITS research, which seeks to enhance the interaction between the ITS as the provider and the learner as the consumer of knowledge, ignoring the crucial role played by the teacher in enhancing the learning in a given context. The limitations of the traditional approach become more visible when educational information is sought to be transmitted across long distances and the need for adaptation to local contexts becomes apparent. This paper argues that the human teacher, as the manager of learning, plays a vital role within the joint cognitive system consisting of the teacher, ITS, learner and learning peers. This role needs to be recognised by ITS designers by through a teacher model. It also suggests that ITS may perhaps best embody the emerging framework of Informing Science.

Predicting corporate bankruptcy using artificial neural networks

An appropriate use of neural computing techniques is to apply them to corporate bankruptcy prediction, where conventional solutions can be hard to obtain. Having said that, choosing an appropriate Artificial Neural Network topology (ANN) for predicting corporate bankruptcy would remain a daunting prospect. The context of the problem is that there are no fixed rules in determining the ANN structure or its parameter values, a large number of ANN topologies may have to be constructed with different structures and parameters before determining an acceptable model. The trial‐and‐error process can be tedious, and the experience of the ANN user in constructing the topologies is invaluable in the search for a good model. Yet, a permanent solution does not exist. This paper identifies a non trivial novel approach for implementing artificial neural networks for the prediction of corporate bankruptcy by applying inter‐connected neural networks. The proposed approach is to produce a neural network architecture that c...

Disc–jet coupling in low-luminosity accreting neutron stars

In outburst, neutron star X-ray binaries produce less powerful jets thanblack holes at a given X-ray luminosity. This has made them moredifficult to study as they fade towards quiescence. To explore whetherneutron stars power jets at low accretion rates (LX ≲1036 erg s-1), we investigate the radio and X-rayproperties of three accreting millisecond X-ray pulsars (IGRJ17511-3057, SAX J1808.4-3658 and IGR J00291+5934) during theiroutbursts in 2015, and of the non-pulsing neutron star Cen X-4 inquiescence (2015) and in outburst (1979). We did not detect the radiocounterpart of IGR J17511-3057 in outburst or of Cen X-4 in quiescence,but did detect IGR J00291+5934 and SAX J1808.4-3658, showing that atleast some neutron stars launch jets at low accretion rates. While theradio and X-ray emission in IGR J00291+5934 seem to be tightlycorrelated, the relationship in SAX J1808.4-3658 is more complicated. Wefind that SAX J1808.4-3658 produces jets during the reflaring tail, andwe explore a toy model to ascertain whether the radio emission could beattributed to the onset of a strong propeller. The lack of a universalradio/X-ray correlation, with different behaviours in different neutronstar systems (with various radio/X-ray correlations; some being radiofaint and others not), points at distinct disc-jet interactions inindividual sources, while always being fainter in the radio band thanblack holes at the same X-ray luminosity.

Rapid variations of polarization in low-mass X-ray binaries

Time-resolved optical and infrared polarimetric observations of black hole and neutron star lowmass X-ray binaries are presented. Data were acquired with the VLT, UKIRT and HIPPO on the SAAO 1.9-m. We find that for some sources in outburst, a rapidly variable component of polarization is evident that is stronger in the redder wavebands.

Selecting the neural network topology for student modelling of prediction of corporate bankruptcy

Neural network topology selection refers to a systematic procedure for selecting between competing models. Naturally, it is regarded as a key aspect in optimisation and replicability of neural network performance. When constructing neural network topologies, it is necessary to determine from the outset the general taxonomy of the neural network architectures to be constructed. The taxonomy considered in this study is the general taxonomy of time‐varying patterns which subsumes many existing architectures in the literature and points to several promising neural network architectures that have yet to be examined. The context of the problem is that choosing the right neural network topology for use in a particular domain such as corporate bankruptcy prediction with optimum generalisation performance is not, in any case, a trivial problem. The results of experiments presented in this paper would serve as a baseline against which to select between two competing architectures.

Resolved, expanding jets in the Galactic black hole candidate XTE J1908+094

Black hole X-ray binaries undergo occasional outbursts caused by changing inner accretion flows. Here we report high angular resolution radio observations of the 2013 outburst of the black hole candidate X-ray binary system XTE J1908+094, using data from the Very Long Baseline Array and European VLBI Network. We show that following a hard-to-soft state transition, we detect moving jet knots that appear asymmetric in morphology and brightness, and expand to become laterally resolved as they move away from the core, along an axis aligned approximately -11. east of north. We initially see only the southern component, whose evolution gives rise to a 15-mJy radio flare and generates the observed radio polarization. This fades and becomes resolved out after 4 days, after which a second component appears to the north, moving in the opposite direction. From the timing of the appearance of the knots relative to the X-ray state transition, a 90. swing of the inferred magnetic field orientation, the asymmetric appearance of the knots, their complex and evolving morphology, and their low speeds, we interpret the knots as working surfaces where the jets impact the surrounding medium. This would imply a substantially denser environment surrounding XTE J1908+094 than has been inferred to exist around the microquasar sources GRS 1915+105 and GRO J1655-40.