<p> <p>&nbsp / </p> </p> <p align="left">IP multicast is an efficient mechanism for simultaneously transmitting bulk data to multiple receivers. Many applications can benefit from multicast, such as audio and videoconferencing, multi-player games, multimedia broadcasting, distance education, and data replication. For either technical or policy reasons, IP multicast still has not yet been deployed in today&rsquo / s Internet. Congestion is one of the most important issues impeding the development and deployment of IP multicast and multicast applications.</p>

In this thesis a hybrid electrical powertrain called the Four Quadrant Transducer (4QT) has been evaluated through different driving simulations, which later resulted in the manufacture of a prototype. The simulation of a 12 metric ton distribution truck showed that the 4QT system can reduce the fuel consumption by approximately 30 % during the FTP75 drive cycle. The reduction in fuel consumption is due to a more optimal control of the combustion engine and regenerative braking of the vehicle. The prototype 4QT has been down scaled from the distribution truck size used in the simulations to a size suitable for a medium sized passenger car. This was done to fit the test rig in the electric machine laboratory. The prototype was tested in the test bench to analyse performances such as efficiency, losses and thermal behaviour. These factors were investigated using both analytical models and the finite element method and later by measurements. The measured results were according to expectations. / I denna doktorsavhandling presenteras ett nytt elhybridsystem för vägfordon benämnt fyrkvadrant omvandlare, "Four Quadrant Transducer (4QT)". Detta system har simulerats under körcykler som t ex FTP75 för att kunna bilda sig en uppfattning om bränsleförbrukningen för hybridsystemet och för att kunna dimensionera elmaskinerna till systemet. En elmaskinprototyp för hybridsystemet har konstruerats och provats i momentvåg. Enligt utförda simuleringar blir besparingen i bränsleförbrukning ca 30% för en tolv tons distributionslastbil utrustad med en 100kW dieselmotor under körcykeln FTP75. Denna minskning av bränsleförbrukning kommer främst från en mera optimal kontroll av förbränningsmotorn samt regenerativ bromsning av fordonet. Den konstruerade prototypen är avsedd för en medelstor bil. Anledningen till att prototypen inte byggdes i en storlek passande för distributionslastbilen var att prototypen skulle passa i testutrustningen i elmaskinlaboratoriet. Prototypen provades i momentvåg för att undersöka verkningsgrad, förluster och termiska prestanda. Resultaten är enligt förväntningarna. / QC 20101014

The thesis is about design and analysis of a PKM with 3 DOF. The new PKM is designed as a machine tool in various applications in manufacturing. The PKM is optimized based on the developed stiffness model. Kinematics and dynamics of the new PKM is also modeled and simulated. / UOIT

This thesis describes research to implement a Bayesian belief network based expert system to solve a real-world diagnostic problem troubleshooting integrated circuit (IC) testing machines. Several models of the IC tester diagnostic problem were developed in belief networks, and one of these models was implemented using Symbolic Probabilistic Inference (SPI). The difficulties and advantages encountered in the process are described in this thesis. It was observed that modelling with interdependencies in belief networks simplified the knowledge engineering task for the IC tester diagnosis problem, by avoiding procedural knowledge and sticking just to diagnostic component's interdependencies. Several general model frameworks evolved through knowledge engineering to capture diagnostic expertise that facilitated expanding and modifying the networks. However, model implementation was restricted to a small portion of the modelling - contact resistance failures - because evaluation of the probability distributions could not be made fast enough to expand the code to a complete model with real-time diagnosis. Further research is recommended to create new methods, or refine existing methods, to speed evaluation of the models created in this research. If this can be done, more complete diagnosis can be achieved. / Graduation date: 1994

Knowledge on cysteine oxidation state and disulfide bond connectivity is of great importance to protein chemistry and 3-D structures. This research is aimed at finding the most relevant features in prediction of cysteines oxidation states and the disulfide bonds connectivity of proteins. Models predicting the oxidation states of cysteines are developed with machine learning techniques such as Support Vector Machines (SVMs) and Associative Neural Networks (ASNNs). A record high prediction accuracy of oxidation state, 95%, is achieved by incorporating the oxidation states of N-terminus cysteines, flanking sequences of cysteines and global information on the protein chain (number of cysteines, length of the chain and amino acids composition of the chain etc.) into the SVM encoding. This is 5% higher than the current methods. This indicates to us that the oxidation states of amino terminal cysteines infer the oxidation states of other cysteines in the same protein chain. Satisfactory prediction results are also obtained with the newer and more inclusive SPX dataset, especially for chains with higher number of cysteines. Compared to literature methods, our approach is a one-step prediction system, which is easier to implement and use. A side by side comparison of SVM and ASNN is conducted. Results indicated that SVM outperform ASNN on this particular problem. For the prediction of correct pairings of cysteines to form disulfide bonds, we first study disulfide connectivity by calculating the local interaction potentials between the flanking sequences of the cysteine pairs. The obtained interaction potential is further adjusted by the coefficients related to the binding motif of enzymes during disulfide formation and also by the linear distance between the cysteine pairs. Finally, maximized weight matching algorithm is applied and performance of the interaction potentials evaluated. Overall prediction accuracy is unsatisfactory compared with the literature. SVM is used to predict the disulfide connectivity with the assumption that oxidation states of cysteines on the protein are known. Information on binding region during disulfide formation, distance between cysteine pairs, global information of the protein chain and the flanking sequences around the cysteine pairs are included in the SVM encoding. Prediction results illustrate the advantage of using possible anchor region information.

A significant problem in many machine learning tasks is that it is time consuming and costly to gather the necessary labeled data for training the learning algorithm to a reasonable level of performance. In reality, it is often the case that a small amount of labeled data is available and that more unlabeled data could be labeled on demand at a cost. If the labeled data is obtained by a process outside of the control of the learner, then the learner is passive. If the learner picks the data to be labeled, then this becomes active learning. This has the advantage that the learner can pick data to gain specific information that will speed up the learning process. Support Vector Machines (SVMs) have many properties that make them attractive to use as a learning algorithm for many real world applications including classification tasks. Some researchers have proposed algorithms for active learning with SVMs, i.e. algorithms for choosing the next unlabeled instance to get label for. Their approach is supervised in nature since they do not consider all unlabeled instances while looking for the next instance. In this thesis, we propose three new algorithms for applying active learning for SVMs in a semi-supervised setting which takes advantage of the presence of all unlabeled points. The suggested approaches might, by reducing the number of experiments needed, yield considerable savings in costly classification problems in the cases when finding the training data for a classifier is expensive.

Support vector machines (SVMs) have been a dominant machine learning technique for more than a decade. The intuitive principle behind SVM training is to find the maximum margin separating hyperplane for a given set of binary labeled training data. Previously, SVMs have been primarily applied to supervised learning problems, where target class labels are provided with the data. Developing unsupervised extensions to SVMs, where no class labels are given, turns out to be a challenging problem. In this dissertation, I propose a principled approach for unsupervised and semi-supervised SVM training by formulating convex relaxations of the natural training criterion: find a (constrained) labeling that would yield an optimal SVM classifier on the resulting labeled training data. This relaxation yields a semidefinite program (SDP) that can be solved in polynomial time. The resulting training procedures can be applied to two-class and multi-class problems, and ultimately to the multivariate case, achieving high quality results in each case. In addition to unsupervised training, I also consider the problem of reducing the outlier sensitivity of standard supervised SVM training. Here I show that a similar convex relaxation can be applied to improve the robustness of SVMs by explicitly suppressing outliers in the training process. The proposed approach can achieve superior results to standard SVMs in the presence of outliers.

Purpose: Vending machines are a component of the food environment that influences dietary choices. Previous vending machine studies have focused on schools and work sites. The purpose of this study was to utilize the Nutrition Environment Measurements Survey-Vending (NEMS-V) online tool to evaluate and rank the nutritional value of the vending environment of a large urban university. Methods: A sample size of 40 vending machines at Georgia State University (GSU) was chosen. A list of products in each machine was recorded and given either a red, yellow or green ranking based on their nutrient content. Finally, the NEMS-V online tool was used to generate a report card for each individual machine and for the entire university. Results: No vending machines were given either the Gold (greater than 50% items ranked green or yellow) or Silver (greater than 40% items ranked green or yellow) ranking. Five machines were given the Bronze level ranking, which meant the machines contained at least 30% yellow or green items. The remaining 35 machines contained less than 30% green or yellow items and were therefore not able to be awarded a ranking. Out of the 40 total machines sampled, less than 30% of them could be ranked and therefore the university could not be given an overall award. Conclusions: The poor nutritional quality of the vending environment at Georgia State University indicates a need for change. Improving the number of vending items from red to yellow or green will offer more variety and more nutritious choices for students.

A significant problem in many machine learning tasks is that it is time consuming and costly to gather the necessary labeled data for training the learning algorithm to a reasonable level of performance. In reality, it is often the case that a small amount of labeled data is available and that more unlabeled data could be labeled on demand at a cost. If the labeled data is obtained by a process outside of the control of the learner, then the learner is passive. If the learner picks the data to be labeled, then this becomes active learning. This has the advantage that the learner can pick data to gain specific information that will speed up the learning process. Support Vector Machines (SVMs) have many properties that make them attractive to use as a learning algorithm for many real world applications including classification tasks. Some researchers have proposed algorithms for active learning with SVMs, i.e. algorithms for choosing the next unlabeled instance to get label for. Their approach is supervised in nature since they do not consider all unlabeled instances while looking for the next instance. In this thesis, we propose three new algorithms for applying active learning for SVMs in a semi-supervised setting which takes advantage of the presence of all unlabeled points. The suggested approaches might, by reducing the number of experiments needed, yield considerable savings in costly classification problems in the cases when finding the training data for a classifier is expensive.

Support vector machines (SVMs) have been a dominant machine learning technique for more than a decade. The intuitive principle behind SVM training is to find the maximum margin separating hyperplane for a given set of binary labeled training data. Previously, SVMs have been primarily applied to supervised learning problems, where target class labels are provided with the data. Developing unsupervised extensions to SVMs, where no class labels are given, turns out to be a challenging problem. In this dissertation, I propose a principled approach for unsupervised and semi-supervised SVM training by formulating convex relaxations of the natural training criterion: find a (constrained) labeling that would yield an optimal SVM classifier on the resulting labeled training data. This relaxation yields a semidefinite program (SDP) that can be solved in polynomial time. The resulting training procedures can be applied to two-class and multi-class problems, and ultimately to the multivariate case, achieving high quality results in each case. In addition to unsupervised training, I also consider the problem of reducing the outlier sensitivity of standard supervised SVM training. Here I show that a similar convex relaxation can be applied to improve the robustness of SVMs by explicitly suppressing outliers in the training process. The proposed approach can achieve superior results to standard SVMs in the presence of outliers.