Multiple instance (MI) learning is a relatively new topic in machine learning. It is concerned with supervised learning but differs from normal supervised learning in two points: (1) it has multiple instances in an example (and there is only one instance in an example in standard supervised learning), and (2) only one class label is observable for all the instances in an example (whereas each instance has its own class label in normal supervised learning). In MI learning there is a common assumption regarding the relationship between the class label of an example and the ``unobservable'' class labels of the instances inside it. This assumption, which is called the ``MI assumption'' in this thesis, states that ``An example is positive if at least one of its instances is positive and negative otherwise''. In this thesis, we first categorize current MI methods into a new framework. According to our analysis, there are two main categories of MI methods, instance-based and metadata-based approaches. Then we propose a new assumption for MI learning, called the ``collective assumption''. Although this assumption has been used in some previous MI methods, it has never been explicitly stated,\footnote{As a matter of fact, for some of these methods, it is actually claimed that they use the standard MI assumption stated above.} and this is the first time that it is formally specified. Using this new assumption we develop new algorithms --- more specifically two instance-based and one metadata-based methods. All of these methods build probabilistic models and thus implement statistical learning algorithms. The exact generative models underlying these methods are explicitly stated and illustrated so that one may clearly understand the situations to which they can best be applied. The empirical results presented in this thesis show that they are competitive on standard benchmark datasets. Finally, we explore some practical applications of MI learning, both existing and new ones. This thesis makes three contributions: a new framework for MI learning, new MI methods based on this framework and experimental results for new applications of MI learning.

Motivated by various challenging real-world applications, such as drug activity prediction and image retrieval, multi-instance (MI) learning has attracted considerable interest in recent years. Compared with standard supervised learning, the MI learning task is more difficult as the label information of each training example is incomplete. Many MI algorithms have been proposed. Some of them are specifically designed for MI problems whereas others have been upgraded or adapted from standard single-instance learning algorithms. Most algorithms have been evaluated on only one or two benchmark datasets, and there is a lack of systematic comparisons of MI learning algorithms. This thesis presents a comprehensive study of MI learning algorithms that aims to compare their performance and find a suitable way to properly address different MI problems. First, it briefly reviews the history of research on MI learning. Then it discusses five general classes of MI approaches that cover a total of 16 MI algorithms. After that, it presents empirical results for these algorithms that were obtained from 15 datasets which involve five different real-world application domains. Finally, some conclusions are drawn from these results: (1) applying suitable standard single-instance learners to MI problems can often generate the best result on the datasets that were tested, (2) algorithms exploiting the standard asymmetric MI assumption do not show significant advantages over approaches using the so-called collective assumption, and (3) different MI approaches are suitable for different application domains, and no MI algorithm works best on all MI problems.

Internet computing enables the construction of large-scale and complex applications by aggregating and sharing computational, data and other resources across institutional boundaries. The agent model can address the ever-increasing challenges of scalability and complexity, driven by the prevalence of Internet computing, by its intrinsic properties of autonomy and reactivity, which support the flexible management of application execution in distributed, open, and dynamic environments. However, the non-deterministic behaviour of autonomous agents leads to a lack of control, which complicates exception management in the system, thus threatening the robustness and reliability of the system, because improperly handled exceptions may cause unexpected system failure and crashes. / In this dissertation, we investigate and develop mechanisms to integrate intrinsic support for concurrency control, exception handling, recoverability, and robustness into multi-agent systems. The research covers agent specification, planning and scheduling, execution, and overall coordination, in order to reduce the impact of environmental uncertainty. Simulation results confirm that our model can improve the robustness and performance of the system, while relieving developers from dealing with the low level complexity of exception handling. / A survey, along with a taxonomy, of existing proposals and approaches for building robust multi-agent systems is provided first. In addition, the merits and limitations of each category are highlighted. / Next, we introduce the ARTS (Agent-Oriented Robust Transactional System) platform which allows agent developers to compose recursively-defined, atomically-handled tasks to specify scoped and hierarchically-organized exception-handling plans for a given goal. ARTS then supports automatic selection, execution, and monitoring of appropriate plans in a systematic way, for both normal and recovery executions. Moreover, we propose multiple-step backtracking, which extends the existing step-by-step plan reversal, to serve as the default exception handling and recovery mechanism in ARTS. This mechanism utilizes previous planning results in determining the response to a failure, and allows a substitutable path to start, prior to, or in parallel with, the compensation process, thus allowing an agent to achieve its goals more directly and efficiently. ARTS helps developers to focus on high-level business logic and relaxes them from considering low-level complexity of exception management. / One of the reasons for the occurrence of exceptions in a multi-agent system is that agents are unable to adhere to their commitments. We propose two scheduling algorithms for minimising such exceptions when commitments are unreliable. The first scheduling algorithm is trust-based scheduling, which incorporates the concept of trust, that is, the probability that an agent will comply with its commitments, along with the constraints of system budget and deadline, to improve the predictability and stability of the schedule. Trust-based scheduling supports the runtime adaptation and evolvement of the schedule by interleaving the processes of evaluation, scheduling, execution, and monitoring in the life cycle of a plan. The second scheduling algorithm is commitment-based scheduling, which focuses on the interaction and coordination protocol among agents, and augments agents with the ability to reason about and manipulate their commitments. Commitment-based scheduling supports the refactoring and parallel execution of commitments to maximize the system's overall robustness and performance. While the first scheduling algorithm needs to be performed by a central coordinator, the second algorithm is designed to be distributed and embedded into the individual agent. / Finally, we discuss the integration of our approaches into Internet-based applications, to build flexible but robust systems. Specifically, we discuss the designs of an adaptive business process management system and of robust scientific workflow scheduling.

Avec la multiplication des services dans un même réseau et les diversités des applications utilisées par les usagers finaux, le trafic transporté est devenu très complexe et dynamique. Le Partage de la Charge Dynamique (PCD) constitue une alternative intéressante pour résoudre cette problématique. Si une paire Source-Destination est connectée par plusieurs chemins, le problème est le suivant : comment distribuer le trafic parmi ces chemins de telle façon qu'une fonction objective soit optimisé. Dans ce cas les chemins sont fixés a priori et la quantité de trafic acheminée sur chaque route est déterminée dynamiquement en fonction de la demande de trafic et de la situation actuelle du réseau. Dans cette thèse nous étudions puis nous proposons plusieurs mécanismes de PCD. Tout d'abord, nous distinguons deux types d'architecture : celles dans lesquelles les ressources sont réservées pour chaque chemin, et celles pour lesquelles aucune réservation n'est effectuée. La simplification faite dans le premier type d'architecture nous permet de proposer l'utilisation d'un nouveau mécanisme pour gérer les chemins. Partant de ce mécanisme, nous définissons un nouvel algorithme de PCD. Concernant la deuxième architecture, nous étudions et comparons plusieurs fonctions objectives. À partir de notre étude, nous proposons un nouvel algorithme distribué permettant d'atteindre l'optimum de ces fonctions objectives. La principale caractéristique de notre algorithme, et son avantage par rapport aux propositions antérieures, est sa capacité d'auto-configuration, dans la mesure où la convergence de l'algorithme est garantie sans aucun besoin de réglage préalable de ses paramètres.

This thesis focuses on the use of computer simulation for modeling risk of a multi-state repairable component. In production processes, maintenance decisions are often made based on uncertain assessment of risk, not only in the probability when a process component goes into a state of failure but also in the cost of lost production and preventive maintenance. In this thesis work, preventive maintenance of a component is modeled and simulated, in order to minimize risk (cost), as: a Markov process with multiple states and fixed transition probabilities, under the assumption that with a sufficient number of states the Markovian property is valid, a non Markov process with two possible states and non-fixed transition probabilities for a periodically decreasing reliability component, and a non Markov process with two possible states and non-fixed transition probabilities for a continuously decreasing reliability component. / Engineering Management

Because of the explosion of digital and online text information, automatic organization of documents has become a very important research area. There are mainly two machine learning approaches to enhance the organization task of the digital documents. One of them is the supervised approach, where pre-defined category labels are assigned to documents based on the likelihood suggested by a training set of labeled documents; and the other one is the unsupervised approach, where there is no need for human intervention or labeled documents at any point in the whole process. In this thesis, we concentrate on the supervised learning task which deals with document classification. One of the most important tasks of information retrieval is to induce classifiers capable of categorizing text documents. The same document can belong to two or more categories and this situation is referred by the term multi-label classification. Multi-label classification domains have been encountered in diverse fields. Most of the existing machine learning techniques which are in multi-label classification domains are extremely expensive since the documents are characterized by an extremely large number of features. In this thesis, we are trying to reduce these computational costs by applying different types of algorithms to the documents which are characterized by large number of features. Another important thing that we deal in this thesis is to have the highest possible accuracy when we have the high computational performance on text document categorization.

Six Ira Sullivan performances were analyzed from studio and live recordings spanning the years 1962 to 1998. Sullivan plays different musical instruments on five of the six selections: trumpet, flute, tenor saxophone, alto saxophone (2 selections), and soprano saxophone. Musical facets considered include phrasing (length/placement), melodic contour, lyricism, harmonic phenomenon, and concept of sound. Common musical threads within Sullivan's improvisations were expected to be found throughout all performances. A call and response dynamic across myriad musical fundamentals such as melody, harmony, and rhythm was found to be present and seems to form a basis for much of Ira Sullivan's improvisations. This and other broad traits common to Sullivan's improvisations are presented herein through analysis.

Truckload shipments constitute a significant portion of the freight transportation industry. In recent years, truckload industry is facing a serious problem of high driver turn over rate. In this research, we present a mathematical model for multi-zone dispatching method to solve this issue. Multi-zone dispatching is a method in which a service area is divided into many zones. Truckload within a zone is carried by local drivers and the truckload between zones is carried by lane drivers. Apart from reducing the driver tour length to a desirable level, the model for multi-zone also contains some unique constraints to address some issues from the perspectives of the company and the customer. The binary integer program is solved by exact methods. As the problem size increases, exact methods fail quickly. Hence, a construction heuristic within tabu search framework is developed to solve the model. Analysis of various parameters concerned is provided to gain better insights of varied aspects of the problem. Computational results for analysis of parameters and comparison of exact and heuristic methods are provided.

Abstract In this literary, we discuss with structure and morphology improvement of GaN epilayer on c-sapphire by multi-step method in molecular-beam epitaxy. Our research is caused for the critical results of defect in GaN epilayer and rough surface morphology. In order to solve these problems we used a novel technique which we called multi-step method. In this thesis, the results of X-ray, SEM, AFM all demonstrated the achievement in our composition. However, we obtained the results of full width of half maxima (FWHM) of (0002) and (10 2) XRD rocking curves with 60~120 arcseconds and 700~ 1200 arcseconds from a series of multi-step samples respectively. Comparing with previous measurement, multi-step method is relatively superior, and the measurement of AFM roughness is under 2 nm from the series of multi-step samples. If we discuss the flat area further, we can get smoother surface which roughness is about 0.4 nm. It is obviously to recognize the flat and rough regions, but in SEM image we made sure that the flat region occupied the greater part of surface. So, in this literary we verified that the method of multi-step can improve the structure and morphology of GaN by molecular-beam epitaxy.

The purpose of this thesis is to develop ring resonators with simple processes and integration. We used loop mirror as a reflector in the semiconductor lasers. In the material, a 1.55£gm symmetric quantum well InGaAsP epi-layer is used to fabricate the lasers. In device design, we designed four kinds of semiconductor lasers by using loop mirror and cleaved facet. The curvature radiuses are 160 and 260£gm that are presented to investigate bending loss and material loss. In the input/output we had an inclined 7 degree to avoid interference. We also designed another two semiconductor lasers by using ring resonator and cleaved facet. Applying the resonance characteristic of ring resonator can achieve wavelength selection and filtering. In fabrication process, we developed new etching technique. The ICP-RIE dry etching and wet etching method were used in the process. Fist we etched half of the total depth by ICP-RIE dry etching. And then the multi-step technique was used to approach the expecting depth. Beside, we had extra deep wet etching process in MMI. Finally, we used the etching solution HBr:HCl:H2O2:H2O =5:4:1:70 to smooth the sidewall and reduce the scattering loss. In device characteristic, we obtained differential quantum efficiency of 20£gW/mA for the 1000£gm straight waveguide laser. We can not observe laser characteristics for the loop mirror laser, partly because of the high loss in bending section.