Graphical model, the marriage between graph theory and probability theory, has been drawing increasing attention because of its many attractive features. In this paper, we consider the problem of learning the structure of graphical model based on observed data through a greedy forward-backward algorithm and with the use of learned model to classify the data into different categories. We establish the graphical model associated with a binary Ising Markov random field. And model selection is implemented by adding and deleting edges between nodes. Our experiments show that: compared with previous methods, the proposed algorithm has better performance in terms of correctness rate and model selection. / text

This thesis is concerned with the utilisation of spatial information in processing of high-frequency sidescan SONAR imagery, and particularly in how such information can be used in developing techniques to assist in mapping functions. Survey applications aim to generate maps of the seabed, but are time consuming and expensive; automatic processing is required to improve efficiency. Current techniques have had some success, but utilise little of the available spatial information. Previously, inclusion of such knowledge was prohibitively expensive; recent improvements in numerical simulations techniques has reduced the costs involved. This thesis attempts to exploit these improvements into a method for including spatial information in SONAR processing and in general to image and signal analysis. Bayesian techniques for inclusion of prior knowledge and structuring complex problems are developed and applied to problems of texture segmentation, object detection and parameter extraction. It is shown through experiments on groundtruth and real datasets that the inclusion of spatial context can be very effective in improving poor techniques or, conversely in allowing simpler techniques to be used with the same objective outcome (with obvious computational advantages). The thesis also considers some of the implementation problems with the techniques used, and develops simple modifications to improve common algorithms.

published_or_final_version / Electrical and Electronic Engineering / Master / Master of Philosophy

The Baum-Welch algorithm for training hidden Markov models (HMMs) requires model topology and initial parameters to be specifed, and iteratively improves the model parameters. Sometimes prior knowledge of the process being modeled allows such specifcation, but often this knowledge is unavailable. Experimentation and guessing are resorted to. Techniques for discovering the model structure from observation data exist but their use is not commonplace. We propose a state split-ting approach to structure discovery, where states are split based on two heuristics: within-state autocorrelation and a measure of Markov violation in the state path. Statistical hypothesis testing is used to decide which states to split, providing a natural termination criterion and taking into account the number of observations assigned to each state, splitting states only when the data demands it. / Thesis (M.A.)-University of KwaZulu-Natal, Durban, 2009.

No description available.

No description available.

No description available.

This thesis examines informal and corruptive activities agents may pursue within organisations. Chapter 1 is a brief introduction to the general theme and the related literature. Chapter 2 develops a simple theory of non-monetary collusion, where agents collude by exchanging favours. It examines the optimal use of supervisory information in a simple hierarchy under potential collusion. It is shown that when only the supervisor's information about the agent is used, collusion does not arise, since favours can not be exchanged. Secondly, it is analysed whether the agent's information about his superior should be used. In this case collusion is possible, and there is an interesting trade-off between the benefits of using additional information and the costs of collusion. It is then shown that sometimes the principal may be better off when using less than all available information. Chapter 3 considers task assignment and whistle-blowing as measures a principal may use to break collusion. The principal's response to potential collusion is to allocate less time to monitoring, and he breaks collusion with money. It is shown that the principal may also break collusion by hiring a third worker, and the decision how to break collusion optimally is endogenously determined. Breaking collusion by task assignment is costly, and therefore we consider whistle-blowing as a collusion breaking device. It provides the principal strictly higher welfare than the collusion-proof solution. It is also shown that under reasonable conditions, the collusion-free outcome will be achieved with no further cost. Chapter 4 develops a model of blackmail, where a piece of information an agent prefers to keep private may facilitate blackmail when another agent, namely a blackmailer, threatens to reveal that information. The crucial feature of the blackmail game is the commitment problem from the blackmailer's side. The blackmailer can not commit not to come back in future to demand more despite the payments received in the past. The chapter outlines conditions under which successful extortion may arise, and shows that there is a unique Markov Perfect Equilibrium, which gives a precise prediction how much money the blackmailer is able to extort from the victim. It is also shown that the blackmailer receives a blackmail premium that compensates the blackmailer for not taking money from the victim and revealing information anyway.

This dissertation considers the random access process in the Medium Access Control (MAC) of communications system. New MAC models are developed to improve the performance of random access based systems. The first contribution is the introduction of a general multichannel random access model with a variable radix. This model is general and can be applied to many existing MAC protocols that utilize random access. It is shown that using the standard Binary Exponential Backoff (BEB) to resolve collisions is not always the best choice. By adjusting the radix, contention efficiency can be improved significantly. The analytical results obtained are confirmed by simulation. The second contribution is the investigation of the variable radix backoff strategy with the contention-based bandwidth request (BW-REQ) mechanism in IEEE 802.16 systems. An analytical model of the BW-REQ procedure is presented which includes a variable radix in the backoff process. Analytical results are presented which show that the variable radix can easily be adjusted to the number of users and the available resources to enhance the efficiency of the Random Access Channel in the uplink subframe. Simulations results are presented to confirm the theory. The third contribution is the development of a reliable Quality of Service (QoS) mechanism for random access systems. The available resources are quantitatively categorized to provide differential services to two classes of users. The model is extended to employ a variable radix strategy. Results show that this strategy can be used in combination with differential services to provide an efficient QoS technique for random access. The fourth contribution is an optimized packet-based finite state Markov chain (FSMC) model for the physical channel. This model employs an equal average fade range duration (AFRD) strategy to partition the signal-to-noise ratio (SNR). The Nakagami-m fading channel model is used as it can span a wide range of fading conditions. The accuracy of the analytical results is confirmed by simulation. A cross-layer Markov model encompassing the FSMC model and a general multichannel random access model is introduced. Finally, a simulation toolbox using object oriented programming is presented. It was used to accurately simulate the models developed in this dissertation. This toolbox is general and can be used for a wide range of MAC models. / Graduate

Designing systems with the ability to make optimal decisions under uncertainty is one of the goals of artificial intelligence. However, in many applications the design of optimal planners is complicated due to imprecise inputs and uncertain outputs resulting from stochastic dynamics. Partially Observable Markov Decision Processes (POMDPs) provide a rich mathematical framework to model these kinds of problems. However, the high computational demand of solution methods for POMDPs is a drawback for applying them in practice. / In this thesis, we present a two-fold approach for improving the tractability of POMDP planning. First, we focus on designing good heuristics for POMDP approximation algorithms. We aim to scale up the efficiency of a class of POMDP approximations called point-based planning methods by designing a good planning space. We study the effect of three properties of reachable belief state points that may influence the performance of point-based approximation methods. Second, we investigate approaches to designing good controllers using an alternative representation of systems with partial observability called Predictive State Representation (PSR). This part of the thesis advocates the usefulness and practicality of PSRs in planning under uncertainty. We also attempt to move some useful characteristics of the PSR model, which has a predictive view of the world, to the POMDP model, which has a probabilistic view of the hidden states of the world. We propose a planning algorithm motivated by the connections between the two models.