Wireless ad hoc networks are suitable and sometimes the only solution for several applications. Many applications, particularly those in military and critical civilian domains (such as battlefield surveillance and emergency rescue) require that ad hoc networks be secure and stable. In fact, security is one of the main barriers to the extensive use of ad hoc networks in many operations. The primary objective of this dissertation is to propose protocols which will protect ad hoc networks from wormhole attacks - one of the most devastating security attacks - and to improve network stability. Protocols that depend solely on cryptography techniques such as authentication and encryption can prevent/detect several types of security attacks; however, they will not be able to detect or prevent a wormhole attack. This attack on routing in ad hoc networks is also considered to be the main threat against neighborhood discovery protocols. Most of the proposed mechanisms designed to defend against this type of attack are based on location information or time measurements, or require additional hardware or a central entity. Other protocols that relied on connectivity or neighborhood information cannot successfully detect all of the various types and cases of wormhole attacks. In the first part of this dissertation, we present a simple, yet effective protocol to detect wormhole attacks along routes in ad hoc networks. The protocol is evaluated using analysis and simulations. In the second part, we present a secure neighbor creation protocol that can securely discover the neighbors of a node in ad hoc networks, and detect and remove wormhole links, if they exist. The proposed protocols do not require any location information, time synchronization, or special hardware to detect wormhole attacks. To the best of our knowledge, this is the first protocol that makes use of cooperation rules between honest nodes. Use of such rules will reduce the overhead associated with the number of checks to be performed in order to detect wormholes and to create a secure neighborhood. This is also the first protocol, to our knowledge, that addresses the complete removal of bogus links without removing legal links.

The performance of mobile ad-hoc networks is normally studied via simulation over a fixed time horizon using a steady-state type of statistical analysis procedure. However, due to the dynamic nature of the network topology such an approach may be inappropriate in many cases as the network may spend most of the time in a transient or nonstationary state. The objective of this dissertation is to develop a performance modeling framework and detailed techniques for analyzing the time varying performance of mobile ad-hoc networks. Our approach is a performance modeling tool for queueing analysis using a hybrid of discrete event simulation and numerical method techniques. Network queues are modeled using fluid-flow based differential equations which can be solved with any standard numerical integration methods, while node connectivity that represents topology changes is incorporated into the model using either discrete event simulation techniques or stochastic modeling of adjacency matrix elements. The hybrid fluid-based approach is believed to be an alternative that can resolve certain issues in current simulators and provide flexibility in modeling a more complex network by integrating additional features of nonstationary effect to add higher level of fidelity into the proposed model. Numerical and simulation experiments show that the new approach can provide reasonably accurate results without sacrificing a large amount of computational resources.

To form a cooperative multiagent team, autonomous agents are required to harmonize activities and make the best use of exclusive resources to achieve their common goal. In addition, to handle uncertainty and quickly respond to external environmental events, they should share knowledge and sensor in formation. Unlike small team coordination, agents in scalable team must limit the amount of their communications while maximizing team performance. Communication decisions are critical to scalable-team coordination because agents should target their communications, but these decisions cannot be supported by a precise model or by complete team knowledge. The hypothesis of my thesis is: local routing of tokens encapsulating discrete elements of control, based only on decentralized local probability decision models, will lead to efficient scalable coordination with several hundreds of agents. In my research, coordination controls including all domain knowledge, tasks and exclusive resources are encapsulated into tokens. By passing tokens around, agents transfer team controls encapsulated in the tokens. The team benefits when a token is passed to an agent who can make use of it, but communications incur costs. Hence, no single agent has sole responsible over any shared decision. The key problem lies in how agents make the correct decisions to target communications and pass tokens so that they will potentially benefit the team most when considering communication costs. My research on token-based coordination algorithm starts from the investigation of random walk of token movement. I found a little increase of the probabilities that agents make the right decision to pass a token, the overall efficiency of the token movement could be greatly enhanced. Moreover, if token movements are modeled as a Markov chain, I found that the efficiency of passing tokens could be significantly varied based on different network topologies. My token-based algorithm starts at the investigation of each single decision theoretic agents. Although under the uncertainties that exist in large multiagent teams, agents cannot act optimal, it is still feasible to build a probability model for each agents to rationally pass tokens. Specifically, this decision only allow agent to pass tokens over an associate network where only a few of team members are considered as token receiver. My proposed algorithm will build each agent's individual decision model based on all of its previously received tokens. This model will not require the complete knowledge of the team. The key idea is that I will make use of the domain relationships between pairs of coordination controls. Previously received tokens will help the receiver to infer whether the sender could benefit the team if a related token is received. Therefore, each token is used to improve the routing of other tokens, leading to a dramatic performance improvement when more tokens are added. By exploring the relationships between different types of coordination controls, an integrated coordination algorithm will be built, and an improvement of one aspect of coordination will enhance the performance of the others.

Recently, several philosophical and computational approaches to causality have used an interventionist framework to clarify the concept of causality [Spirtes et al., 2000, Pearl, 2000, Woodward, 2005]. The characteristic feature of the interventionist approach is that causal models are potentially useful in predicting the effects of manipulations. One of the main motivations of such an undertaking comes from humans, who seem to create sophisticated mental causal models that they use to achieve their goals by manipulating the world. Several algorithms have been developed to learn static causal models from data that can be used to predict the effects of interventions [e.g., Spirtes et al., 2000]. However, Dash [2003, 2005] argued that when such equilibrium models do not satisfy what he calls the Equilibration-Manipulation Commutability (EMC) condition, causal reasoning with these models will be incorrect, making dynamic models indispensable. It is shown that existing approaches to learning dynamic models [e.g., Granger, 1969, Swanson and Granger, 1997] are unsatisfactory, because they do not perform a necessary search for hidden variables. The main contribution of this dissertation is, to the best of my knowledge, the first provably correct learning algorithm that discovers dynamic causal models from data, which can then be used for causal reasoning even if the EMC condition is violated. The representation that is used for dynamic causal models is called Difference-Based Causal Models (DBCMs) and is based on Iwasaki and Simon [1994]. A comparison will be made to other approaches and the algorithm, called DBCM Learner, is empirically tested by learning physical systems from artificially generated data. The approach is also used to gain insights into the intricate workings of the brain by learning DBCMs from EEG data and MEG data.

Web-based tagging systems, which include social bookmarking systems such as Delicious, have become increasingly popular. These systems allow participants to annotate or tag web resources. This research examined the use of social annotations to improve the quality of web searches. The research involved three components. First, social annotations were used to index resources. Two annotation-based indexing methods were proposed: annotation based indexing and full text with annotation indexing. Second, social annotations were used to improve search result ranking. Six annotation based ranking methods were proposed: Popularity Count, Propagate Popularity Count, Query Weighted Popularity Count, Query Weighted Propagate Popularity Count, Match Tag Count and Normalized Match Tag Count. Third, social annotations were used to both index and rank resources. The result from the first experiment suggested that both static feature and similarity feature should be considered when using social annotations to re-rank search result. The result of the second experiment showed that using only annotation as an index of resources may not be a good idea. Since social Annotations could be viewed as a high level concept of the content, combining them to the content of resource could add some more important concepts to the resources. Last but not least, the result from the third experiment confirmed that the combination of using social annotations to rank the search result and using social annotations as resource index augmentation provided a promising rank of search results. It showed that social annotations could benefit web search.

While there are many studies on collaborative or guided scientific inquiry in real, virtual, and simulated environments, there are few that study the interplay between the design of the simulation and the user interface. The main research aim was to decompose the simulation and user interface into the design parameters that influence attention, curiosity, inquiry, and learning of scientific material and acts of creation for children. The research design investigates what tools support independent exploration of a space, enhance deep learning, and motivate scientific or creative inquiry. A major interest is in the role that ecological context plays in the perception of spatial information. None of the prior work on learning in virtual environments considered a child-centric computer interaction framing, independent of pedagogy and focused on the impact of user interface parameters, such as image quality and navigational freedom. A major contribution of this research is the construction of the Virtual Trillium Trail, as it represents one square mile of biologically accurate scientific plot study data. It is a virtual environment based on statistical data visualization, not fantasy. It allowed for a highly realistic simulation and scientifically true-to-life visualization, as well as for a planned orthogonal contrast with exceptionally high internal validity in both system and statistical research design. Of critical importance is evidence in the pilot study, that virtual reality field trips for students may be used to prime before and to reinforce after a real field trip. This research also showed transfer effects on in-situ learning activity, in both directions. Thus, supports the claim that virtual environments may augment educational practices, not replace them, to maximize the overall learning impact. The other large contribution was in the activity analysis of the real field trip, where the Salamander Effect is observed as an environmental event, which opened a Teachable Moment event for the teacher, and which was then translated into a system design feature, a Salient Event in the user interface. A main part of this research is the importance of such events, as ways to support intrinsic learning activity, and leverage episodic memory. The main empirical contribution to the design of educational virtual environments was produced by the 2 x 2 ANOVA with the factors of Visual Fidelity and Navigational Freedom, set to high and low levels, and the evidence of different effects on Knowledge Gained. The tool has an impact on intrinsic learning, which is measured here by a pretest and a post-test on facts and concepts. A two-factor analysis of variance showed a significant effect of Visual Fidelity on Knowledge Gained, F(1,60) = 10.54, p = 0.0019. High Visual Fidelity condition has a greater impact on Knowledge Gained (M=30.95, SD =14.76), than Low Visual Fidelity condition (M=19.99, SD = 13.39). Photorealistic has a stronger impact on learning than cartoon versions. There was significant interaction between Visual Fidelity and Navigational Freedom, F(1,60) = 4.85, p = 0.0315, with the largest impact in the combined conditions of High Visual Fidelity and High Navigational Freedom on Knowledge Gained (M=37.44, SD = 13.88). Thus, photorealistic, free navigation virtual environments double learning, when compared to cartoon versions, ceteris paribus. The next major contribution to the design of the user interface in educational virtual environments is the design and use of Salient Events as components to augment the virtual environment and to facilitate intrinsic inquiry into facts and concepts. A two factor analysis of variance showed a significant effect of Visual Fidelity on Salient Event counts, F(1,60) = 4.35, p = 0.00413. High Visual Fidelity condition has a greater impact on Salient Event counts, (Μ = 14.46, SD = 6), than Low Visual Fidelity condition, (Μ =11.31, SD = 6.37). Using High Visual Fidelity with High Navigational Freedom (showing a strong trend of F(1,60) = 3.23, p = 0.0773) to increase Salient Event counts are critical design features for educational virtual environments, especially since Salient Events are moderately positively correlated with Knowledge Gained (r = 0.455, N = 64, p = 0.000). Emotional, affective, aesthetic, and subjective attitudes were investigated in the post-experience assessment of the main study on system and learning experience. Total Attitude is strongly positively and significantly correlated with Awe and Wonder (r = 0.727, N = 64, p = 0.000). Also important is the strong, positive, and significant correlation of Beauty with Awe and Wonder (r = 0.506, N = 64, p = 0.000). And the only significant subjective emotion or attitude variable correlated to Knowledge Gained, was Awe and Wonder with a slightly positive statistic: (r = 0.273, N = 64, p = 0.000). Future research will investigate the complexity and causality of such interactions between the childs mental model, the virtual environment, and the user interface in the form of regression equations, partial differential equations, and Markov models.

Ensemble approaches have been shown to enhance classification by combining the outputs from a set of voting classifiers. Diversity in error patterns among base classifiers promotes ensemble performance. Multi-task learning is an important characteristic for Neural Network classifiers. Introducing a secondary output unit that receives different training signals for base networks in an ensemble can effectively promote diversity and improve ensemble performance. Here a Competitive Learning Neural Network Ensemble is proposed where a secondary output unit predicts the classification performance of the primary output unit in each base network. The networks compete with each other on the basis of classification performance and partition the stimulus space. The secondary units adaptively receive different training signals depending on the competition. As the result, each base network develops ¡°preference¡± over different regions of the stimulus space as indicated by their secondary unit outputs. To form an ensemble decision, all base networks¡¯ primary unit outputs are combined and weighted according to the secondary unit outputs. The effectiveness of the proposed approach is demonstrated with the experiments on one real-world and four artificial classification problems.

An ontology mapping neural network (OMNN) is proposed in order to learn and infer correspondences among ontologies. It extends the Identical Elements Neural Network (IENN)s ability to represent and map complex relationships. The learning dynamics of simultaneous (interlaced) training of similar tasks interact at the shared connections of the networks. The output of one network in response to a stimulus to another network can be interpreted as an analogical mapping. In a similar fashion, the networks can be explicitly trained to map specific items in one domain to specific items in another domain. Representation layer helps the network learn relationship mapping with direct training method. The OMNN approach is tested on family tree test cases. Node mapping, relationship mapping, unequal structure mapping, and scalability test are performed. Results show that OMNN is able to learn and infer correspondences in tree-like structures. Furthermore, OMNN is applied to several OAEI benchmark test cases to test its performance on ontology mapping. Results show that OMNN approach is competitive to the top performing systems that participated in OAEI 2009.

This thesis explores four research areas that are examined using DNS traffic analysis. The tools used for this analysis are presented first. The four topics examined are domain mapping, response time of anti-phishing block lists to find the phishing sites, automated identification of malicious fast-flux hosting domains, and identification of distributed denial of service attacks. The first three approaches yielded successful results, and the fourth yields primarily negative lessons for using DNS traffic analysis in such a scenario. Much of the analysis concerns the anti-phishing response time, which has yielded tentative results. It is found that there is significant overlap between the automatically identified fast-flux sites and those sites on the block list. It appears that domains were being put onto the list approximately 11 hours after becoming active, in the median case, which is very nearly the median lifetime of a phishing site. More recently collected data indicates that this result is extremely difficult to verify. While further work is necessary to verify these claims, the initial indication is that finding and listing phishing sites is the bottleneck in propagating data to protect consumers from malicious phishing sites.

This research explores the role of affect in the domain of human wayfinding by asking if increased motivation will alter the performance across various routes of increasing complexity. Participants were asked to perform certain navigation tasks within an indoor Virtual Reality (VR) environment under either motivated and not-motivated instructions. After being taught to navigate along simple and complex routes, participants were tested on both the previously learned routes and new routes that could be implicitly derived from the prior spatial knowledge. Finally, participants were tested on their ability to follow schematized instructions to explore familiar and unfamiliar areas in the VR environment. Performance of the various spatial tasks across the motivated and control groups indicated that motivation improved performance in all but the most complex conditions. Results of the empirical study were used to create a theoretical model that accounts for the influence of affect on the access of route knowledge. Results of the research suggest the importance of including past knowledge and affect of the traveler as components of future wayfinding systems.