There is a high degree of uncertainty regarding the nature of traffic on future integrated service networks. This uncertainty motivates the use of adaptive resource allocation policies that can take advantage of the statistical fluctuations in the traffic demands. The adaptive control mechanisms must be 'lightweight', in terms of their overheads, and scale to potentially large networks with many traffic flows. Adaptive routing is one form of adaptive resource allocation, and this thesis considers the application of Stochastic Learning Automata (SLA) for distributed, lightweight adaptive routing in future integrated service communication networks. The thesis begins with a broad critical review of the use of Artificial Intelligence (AI) techniques applied to the control of communication networks. Detailed simulation models of integrated service networks are then constructed, and learning automata based routing is compared with traditional techniques on large scale networks. Learning automata are examined for the 'Quality-of-Service' (QoS) routing problem in realistic network topologies, where flows may be routed in the network subject to multiple QoS metrics, such as bandwidth and delay. It is found that learning automata based routing gives considerable blocking probability improvements over shortest path routing, despite only using local connectivity information and a simple probabilistic updating strategy. Furthermore, automata are considered for routing in more complex environments spanning issues such as multi-rate traffic, trunk reservation, routing over multiple domains, routing in high bandwidth-delay product networks and the use of learning automata as a background learning process. Automata are also examined for routing of both 'real-time' and 'non-real-time' traffics in an integrated traffic environment, where the non-real-time traffic has access to the bandwidth 'left over' by the real-time traffic. It is found that adopting learning automata for the routing of the real-time traffic may improve the performance to both real and non-real-time traffics under certain conditions. In addition, it is found that one set of learning automata may route both traffic types satisfactorily. Automata are considered for the routing of multicast connections in receiver-oriented, dynamic environments, where receivers may join and leave the multicast sessions dynamically. Automata are shown to be able to minimise the average delay or the total cost of the resulting trees using the appropriate feedback from the environment. Automata provide a distributed solution to the dynamic multicast problem, requiring purely local connectivity information and a simple updating strategy. Finally, automata are considered for the routing of multicast connections that require QoS guarantees, again in receiver-oriented dynamic environments. It is found that the distributed application of learning automata leads to considerably lower blocking probabilities than a shortest path tree approach, due to a combination of load balancing and minimum cost behaviour.

Indiana University-Purdue University Indianapolis (IUPUI) / Multi-criteria decision making (MCDM) methods have evolved over the past several decades. In today’s world with rapidly growing industries, MCDM has proven to be significant in many application areas. In this study, a decision-making model is devised using reinforcement learning to carry out multi-criteria optimization problems. Learning automata algorithm is used to identify an optimal solution in the presence of single and multiple environments (criteria) using pareto optimality. The application of this model is also discussed, where the model provides an optimal solution to the food, energy, and water systems (FEWS) problem.

One of today's most serious social, economical and environmental problems is traffic congestion. In addition to the financial cost of the problem, the number of traffic related injuries and casualties is very high. A recently considered approach to increase safety while reducing congestion and improving driving conditions is Automated Highway Systems (AHS). The AHS will evolve from the present highway system to an intelligent vehicle/highway system that will incorporate communication, vehicle control and traffic management techniques to provide safe, fast and more efficient surface transportation. A key factor in AHS deployment is intelligent vehicle control. While the technology to safely maneuver the vehicles exists, the problem of making intelligent decisions to improve a single vehicle's travel time and safety while optimizing the overall traffic flow is still a stumbling block. We propose an artificial intelligence technique called stochastic learning automata to design an intelligent vehicle path controller. Using the information obtained by on-board sensors and local communication modules, two automata are capable of learning the best possible (lateral and longitudinal) actions to avoid collisions. This learning method is capable of adapting to the automata environment resulting from unmodeled physical environment. Simulations for simultaneous lateral and longitudinal control of an autonomous vehicle provide encouraging results. Although the learning approach taken is capable of providing a safe decision, optimization of the overall traffic flow is also possible by studying the interaction of the vehicles. The design of the adaptive vehicle path planner based on local information is then carried onto the interaction of multiple intelligent vehicles. By analyzing the situations consisting of conflicting desired vehicle paths, we extend our design by additional decision structures. The analysis of the situations and the design of the additional structures are made possible by the study of the interacting reward-penalty mechanisms in individual vehicles. The definition of the physical environment of a vehicle as a series of discrete state transitions associated with a "stationary automata environment" is the key to this analysis and to the design of the intelligent vehicle path controller. This work was supported in part by the Center for Transportation Research and Virginia DOT under Smart Road project, by General Motors ITS Fellowship program, and by Naval Research Laboratory under grant no. N000114-93-1-G022. / Ph. D.

The field of Learning Automata (LA) has been studied and analyzed extensively for more than four decades; however, almost all the papers have concentrated on the LA working in Environments that have a finite number of actions. This is a well-established model of computation, and expedient, epsilon-optimal and absolutely expedient machines have been designed for stationary and non-stationary Environments. There are only a few papers which deal with Environments possessing an infinite number of actions. These papers assume a well-defined and rather simple uni-modal functional form, like the Gaussian function, for the Environment's infinite reward probabilities. This thesis pioneers the concept and presents a series of continuous action LA (CALA) algorithms that do not require the function of the Environment's infinite reward probabilities to obey a well-established uni-modal functional form. Instead, this function can be, but not limited to, a multi-modal function as long as it satisfies some weak constraints. Moreover, as our discussion evolves, the constraints are further relaxed. In all these cases, we demonstrate that the underlying machines converge in an epsilon-optimal manner to the optimal action of an infinite action set. Based on the CALA algorithms proposed, we report a global maximum search algorithm, which can find the maximum points of a real-valued function by sampling the function's values that could be contaminated by noise. This thesis also investigates the performance limit of the action-taking scheme, sampling actions based on probability density functions, which is used by all currently available CALA algorithms. In more details, given a reward function, we define an index of the function which is the least upper bound of the performance that a CALA algorithm can possibly achieve. Besides, we also report a CALA algorithm that meets this upper bound in an epsilon-optimal manner. By investigating the problem from a different perspective, we argue that the algorithms proposed are closely related to the family of “Stochastic Point Location” problems involving either discretized steps or d-ary parallel machines. The thesis includes the detailed proofs of the assertions and highlights the niche contributions within the broader theory of learning. To the best of our knowledge, there are no comparable results reported in the literature.

Grouping is a vital precursor to object recognition. The complexity of the object recognition process can be reduced to a large extent by using a frontend grouping process. In this dissertation, a grouping framework based on spectral methods for graphs is used. The objects are segmented from the background by means of an associated learning process that decides on the relative importance of the basic salient relationships such as proximity, parallelism, continuity, junctions and common region. While much of the previous research has been focussed on using simple relationships like similarity, proximity, continuity and junctions, this work differenciates itself by using all the relationships listed above. The parameters of the grouping process is cast as probabilistic specifications of Bayesian networks that need to be learned: the learning is accomplished by a team of stochastic learning automata. One of the stages in the grouping process is graph partitioning. There are a variety of cut measures based on which partitioning can be obtained and different measures give different partitioning results. This work looks at three popular cut measures, namely the minimum, average and normalized. Theoretical and empirical insight into the nature of these partitioning measures in terms of the underlying image statistics are provided. In particular, the questions addressed are as follows: For what kinds of image statistics would optimizing a measure, irrespective of the particular algorithm used, result in correct partitioning? Are the quality of the groups significantly different for each cut measure? Are there classes of images for which grouping by partitioning is not suitable? Does recursive bi-partitioning strategy separate out groups corresponding to K objects from each other? The major conclusion is that optimization of none of the above three measures is guaranteed to result in the correct partitioning of K objects, in the strict stochastic order sense, for all image statistics. Qualitatively speaking, under very restrictive conditions when the average inter-object feature affinity is very weak when compared to the average intra-object feature affinity, the minimum cut measure is optimal. The average cut measure is optimal for graphs whose partition width is less than the mode of distribution of all possible partition widths. The normalized cut measure is optimal for a more restrictive subclass of graphs whose partition width is less than the mode of the partition width distributions and the strength of inter-object links is six times less than the intra-object links. The learning framework described in the first part of the work is used to empirically evaluate the cut measures. Rigorous empirical evaluation on 100 real images indicates that in practice, the quality of the groups generated using minimum or average or normalized cuts are statistically equivalent for object recognition, i.e. the best, the mean, and the variation of the qualities are statistically equivalent. Another conclusion is that for certain image classes, such as aerial and scenes with man-made objects in man-made surroundings, the performance of grouping by partitioning is the worst, irrespective of the cut measure.

No description available.

No description available.

<p>Multi-criteria decision making (MCDM) methods have evolved over the past several decades. In today’s world with rapidly growing industries, MCDM has proven to be significant in many application areas. In this study, a decision-making model is devised using reinforcement learning to carry out multi-criteria optimization problems. Learning automata algorithm is used to identify an optimal solution in the presence of single and multiple environments (criteria) using pareto optimality. The application of this model is also discussed, where the model provides an optimal solution to the food, energy, and water systems (FEWS) problem.</p>

The PhD thesis describes control design of active magnetic bearing. Active magnetic bearing is nonlinear unstable system. This means it is not possible to use classic methods of control design for linear time invariant systems. Also methods of nonlinear control design are not universal and theirs application is not easy task. The thesis describes usage of simple nonlinear compensation which linearizes response of active magnetic bearing and allows usage of classic methods of control design for linear time invariant systems. It is shown that CARLA method can significantly improve parameters of designed controller. First part of thesis describes derivation of model of controlled active magnetic bearing and nonlinear compensation which linearizes response of controlled active magnetic bearing on input signal. Following part contains description of methods of state control design methods, selected methods of robust control design and most common methods of artificial intelligence used for control design and implementation. Next part describes hardware of used experimental device and its parameters. It also contains experimental derivation of model of electromagnetic force because the parameters are not available from manufacturer. Last part describes control design of active magnetic bearing. Several different approaches are described here. The approaches vary from completely experimental approach, through using Ziegler-Nichols method, state control design to methods for robust control design. During design is heavily used CARLA method which is very suitable for usage for online learning in real controller due its principle.

No description available.