Catálogo de publicaciones - libros

In this paper we show how to capture dialogue games in Defeasible Logic. We argue that Defeasible Logic is a natural candidate and general representation formalism to capture dialogue games even with requirements more complex than existing formalisms for this kind of games. We parse the dialogue into defeasible rules with time of the dialogue as time of the rule. As the dialogue evolves we allow an agent to upgrade the strength of unchallenged rules. The proof procedures of [1] are used to determine the winner of a dialogue game.

Belief change is concerned with modelling the way in which an idealised (rational) reasoner maintains their beliefs and the way in which those beliefs are modified as the reasoner acquires new information. The AGM [1,3,5] framework is the most widely cited belief change methodology in the literature. It models the reasoner’s belief state as a set of sentences that is logically closed under deduction and provides for three belief change operations: expansion, contraction and revision. Each of the AGM belief change operations is motivated by that are formalised by way of .

Pagnucco [10] formalised a way of implementing the AGM belief change operations using a technique involving in order to improve computational efficiency. This technique exploits the construction for AGM belief change by Gärdenfors and Makinson [4] by introducing the notion of a . It has a number of significant features: (a) the belief change operators constructed satisfy the AGM postulates; (b) when compilation has been effected only subsumption checking and some simple syntactic manipulation is required in order to contract (or revise) the reasoner’s belief state.

This paper details updated results concerning an implementation of a Multiple Classification Ripple Down Rules (MCRDR) system which can be used to provide quality Decision Support Services to pharmacists practicing medication reviews (MRs), particularly for high risk patients. The system was trained on 126 genuine cases by an expert in the field; over the course of 19 hours the system had learned 268 rules and was considered to encompass over 80% of the domain. Furthermore, the system was found able to improve the quality and consistency of the medication review reports produced, as it was shown that there was a high incidence of missed classifications under normal conditions, which were repaired by the system automatically. However, shortcomings were identified including an inability to handle absent data, and shortcomings concerning standardization in the domain, proposals to solve these shortcomings are included.

This paper describes a mathematical framework for studying a nonlinear feedback control. The fuzzy control discussed here is the nonlinear feedback control in which the feedback laws are determined by IF-THEN type fuzzy production rules through approximate reasoning introduced by Nakamori. To prove existence of optimal control, we applied compactness of a set of membership functions in space and continuity of the approximate reasoning, and prepared some propositions concerning approximate reasoning of Nakamori model. By considering fuzzy optimal control problems as problems of finding the minimum (maximum) value of the integral cost (benefit) function on an appropriate set of membership functions, the existence of fuzzy optimal control is shown.

Although Computed Tomography (CT) is a mature discipline, the development of techniques that will further reduce radiation dose are still essential. This paper makes steps towards projection andreconstruction methods which aim to assist in the reduction of this dosage, by studying the way noise propagates from projection space to image space. Inference methods Maximum Likelihood Estimation (MLE), Akaike’s Information Criterion (AIC) and Minimum Message Length (MML) are used to obtain accurate models obtained from minimal data.

Underwater video is increasingly being pursued as a low impact alternative to traditional techniques (such as trawls and dredges) for determining abundance and size frequency of target species. Our researchfocuses on automatically annotating survey scallop video footage using artificial intelligence techniques. We use a multi-layered approach which implements an attention selection process followed by sub-image segmentation and classification. Initial attention selection is performed using the University of Southern California’s (USCs) iLab Neuromorphic Visual Toolkit (iNVT). Once the iNVT has determined regions of potential interest we use image segmentation and feature extraction techniques to produce data suitable for analysis within the Weka machine learning workbench environment.

This paper describes an extension to the Haar Classifier Cascade technique for object detection. Existing Haar Classifier Cascades are binary; the extension adds confidence measurement. This confidence measure was implemented and found to improve accuracy on two object detection problems: face detection and fish detection. For fish detection, the problem of selecting positive training-sample angle-ranges was also considered; results showed that large random variations that result in cascades covering overlapping ranges increases their accuracy.

This paper describes our research in technologies for the management and control of distributed energy resources. An agent-based management and control system is being developed to enable large-scale deployment of distributed energy resources. Local intelligent agents will allow consumers who are connected at low levels in the distribution network to manage their energy requirements and participate in coordination responses to network stimuli. Such responses can be used to reduce the volatility of wholesale electricity prices and assist constrained networks during summer and winter demand peaks. The management and control of very large numbers of distributed energy resources to create aggregated quantities of power can be used to improve the efficiency of the electricity network and market. In our system, the coordination of energy resources is decentralized. Energy resources coordinate each other to realize efficient autonomous matching of supply and demand in large power distribution networks. The information exchange is through indirect (or stigmergic) communications between resource agents and a broker agent. The coordination mechanism is asynchronous and adapts to change in an unsupervised manner, making it intrinsically scalable and robust.

Case adaptation continues to be one of the more difficult aspects of case-based reasoning to automate. This paper looks at several techniques for utilising the implicit knowledge contained in a case base for case adaptation in case-based reasoning systems. The most significant of the techniques proposed are a moderately successful data mining technique and a highly successful artificial neural network technique. Their effectiveness was evaluated on a footwear design problem.

Building classification models plays an important role in DNA mircroarray data analyses. An essential feature of DNA microarray data sets is that the number of input variables (genes) is far greater than the number of samples. As such, most classification schemes employ variable selection or feature selection methods to pre-process DNA microarray data. This paper investigates various aspects of building classification models from microarray data with tree-based classification algorithms by using Partial Least-Squares (PLS) regression as a feature selection method. Experimental results show that the Partial Least-Squares (PLS) regression method is an appropriate feature selection method and tree-based ensemble models are capable of delivering high performance classification models for microarray data.