GYAN: A methodology for rule extraction from artificial neural networks

Artificial neural network (ANN) learning methods provide a robust and non-linear approach to approximating the target function for many classification, regression and clustering problems. ANNs have demonstrated good predictive performance in a wide variety of practical problems. However, there are strong arguments as to why ANNs are not sufficient for the general representation of knowledge. The arguments are the poor comprehensibility of the learned ANN, and the inability to represent explanation structures. The overall objective of this thesis is to address these issues by: (1) explanation of the decision process in ANNs in the form of symbolic rules (predicate rules with variables); and (2) provision of explanatory capability by mapping the general conceptual knowledge that is learned by the neural networks into a knowledge base to be used in a rule-based reasoning system. A multi-stage methodology GYAN is developed and evaluated for the task of extracting knowledge from the trained ANNs. The extracted knowledge is represented in the form of restricted first-order logic rules, and subsequently allows user interaction by interfacing with a knowledge based reasoner. The performance of GYAN is demonstrated using a number of real world and artificial data sets. The empirical results demonstrate that: (1) an equivalent symbolic interpretation is derived describing the overall behaviour of the ANN with high accuracy and fidelity, and (2) a concise explanation is given (in terms of rules, facts and predicates activated in a reasoning episode) as to why a particular instance is being classified into a certain category.

Development and synthesis of innovative approaches to the surgical assessment of articular cartilage degeneration

This research has established, through ultrasound, near infrared spectroscopy and biomechanics experiments, parameters and parametric relationships that can form the framework for quantifying the integrity of the articular cartilage-on-bone laminate, and objectively distinguish between normal/healthy and abnormal/degenerated joint tissue, with a focus on articular cartilage. This has been achieved by: 1. using traditional experimental methods to produce new parameters for cartilage assessment; 2. using novel methodologies to develop new parameters; and 3. investigating the interrelationships between mechanical, structural and molec- ular properties to identify and select those parameters and methodologies that can be used in a future arthroscopic probe based on points 1 and 2. By combining the molecular, micro- and macro-structural characteristics of the tissue with its mechanical properties, we arrive at a set of critical benchmarking parameters for viable and early-stage non-viable cartilage. The interrelationships between these characteristics, examined using a multivariate analysis based on principal components analysis, multiple linear regression and general linear modeling, could then to deter- mine those parameters and relationships which have the potential to be developed into a future clinical device. Specifically, this research has found that the ultrasound and near infrared techniques can subsume the mechanical parameters and combine to characterise the tissue at the molecular, structural and mechanical levels over the full depth of the cartilage matrix. It is the opinion in this thesis that by enabling the determination of the precise area of in uence of a focal defect or disease in the joint, demarcating the boundaries of articular cartilage with dierent levels of degeneration around a focal defect, better surgical decisions that will advance the processes of joint management and treatment will be achieved. Providing the basis for a surgical tool, this research will contribute to the enhancement and quanti�cation of arthroscopic procedures, extending to post- treatment monitoring and as a research tool, will enable a robust method for evaluating developing (particularly focalised) treatments.

Application of micro CT and computation modeling in bone tissue engineering

Computer aided technologies, medical imaging, and rapid prototyping has created new possibilities in biomedical engineering. The systematic variation of scaffold architecture as well as the mineralization inside a scaffold/bone construct can be studied using computer imaging technology and CAD/CAM and micro computed tomography (CT). In this paper, the potential of combining these technologies has been exploited in the study of scaffolds and osteochondral repair. Porosity, surface area per unit volume and the degree of interconnectivity were evaluated through imaging and computer aided manipulation of the scaffold scan data. For the osteochondral model, the spatial distribution and the degree of bone regeneration were evaluated. In this study the versatility of two softwares Mimics (Materialize), CTan and 3D realistic visualization (Skyscan) were assessed, too.

Adapting a generic BEAMnrc model of the BrainLAB m3 micro-multileaf collimator to simulate a local collimation device

This work is focussed on developing a commissioning procedure so that a Monte Carlo model, which uses BEAMnrc’s standard VARMLC component module, can be adapted to match a specific BrainLAB m3 micro-multileaf collimator (μMLC). A set of measurements are recommended, for use as a reference against which the model can be tested and optimised. These include radiochromic film measurements of dose from small and offset fields, as well as measurements of μMLC transmission and interleaf leakage. Simulations and measurements to obtain μMLC scatter factors are shown to be insensitive to relevant model parameters and are therefore not recommended, unless the output of the linear accelerator model is in doubt. Ultimately, this note provides detailed instructions for those intending to optimise a VARMLC model to match the dose delivered by their local BrainLAB m3 μMLC device.

The application of neural networks to induction machine control

Neural networks (NNs) are discussed in connection with their possible use in induction machine drives. The mathematical model of the NN as well as a commonly used learning algorithm is presented. Possible applications of NNs to induction machine control are discussed. A simulation of an NN successfully identifying the nonlinear multivariable model of an induction-machine stator transfer function is presented. Previously published applications are discussed, and some possible future applications are proposed.

Identification and control of induction machines using artificial neural networks

The use of artificial neural networks (ANNs) to identify and control induction machines is proposed. Two systems are presented: a system to adaptively control the stator currents via identification of the electrical dynamics, and a system to adaptively control the rotor speed via identification of the mechanical and current-fed system dynamics. Both systems are inherently adaptive as well as self-commissioning. The current controller is a completely general nonlinear controller which can be used together with any drive algorithm. Various advantages of these control schemes over conventional schemes are cited, and the combined speed and current control scheme is compared with the standard vector control scheme

Identification and control of induction machines using artificial neural networks

This paper proposes the use of artificial neural networks (ANNs) to identify and control an induction machine. Two systems are presented: a system to adaptively control the stator currents via identification of the electrical dynamics; and a system to adaptively control the rotor speed via identification of the mechanical and current-fed system dynamics. Various advantages of these control schemes over other conventional schemes are cited and the performance of the combined speed and current control scheme is compared with that of the standard vector control scheme

Alleviating ‘overfitting’ via genetically-regularised neural network

A hybrid genetic algorithm/scaled conjugate gradient regularisation method is designed to alleviate ANN `over-fitting'. In application to day-ahead load forecasting, the proposed algorithm performs better than early-stopping and Bayesian regularisation, showing promising initial results.

Oxidative potential of logwood and pellet burning particles assessed by a novel profluorescent nitroxide probe

This study reports the potential toxicological impact of particles produced during biomass combustion by an automatic pellet boiler and a traditional logwood stove under various combustion conditions using a novel profluorescent nitroxide probe BPEAnit. This probe is weakly fluorescent, but yields strong fluorescence emission upon radical trapping or redox activity. Samples were collected by bubbling aerosol through an impinger containing BPEAnit solution, followed by fluorescence measurement. The fluorescence of BPEAnit was measured for particles produced during various combustion phases, at the beginning of burning (cold start), stable combustion after refilling with the fuel (warm start) and poor burning conditions. For particles produced by the logwood stove under cold-start conditions significantly higher amounts of reactive species per unit of particulate mass were observed compared to emissions produced during a warm start. In addition, sampling of logwood burning emissions after passing through a thermodenuder at 250oC resulted in an 80-100% reduction of the fluorescence signal of BPEAnit probe, indicating that the majority of reactive species were semivolatile. Moreover, the amount of reactive species showed a strong correlation with the amount of particulate organic material. This indicates the importance of semivolatile organics in particle-related toxicity. Particle emissions from the pellet boiler, although of similar mass concentration, were not observed to lead to an increase in fluorescence signal during any of the combustion phases.