Social network web sites become very popular nowadays. Users can establish connections with other users forming a social network, and quickly share information, photographs, and videos with friends. Malwares called social network worms can send text messages with malicious URLs by employing social engineering techniques. They are trying let users click malicious URL and infect users. Also, it can quickly attack others by infected user accounts in social network. By curiosity, most users click it without validation. This thesis proposes a malicious URL detection method used in Facebook wall, which used heuristic features with high classification property and machine learning algorithm, to predict the safety of URL messages. Experiments show that, the proposed approach can achieve about 96.3% of True Positive Rate, 95.4% of True Negative Rate, and 95.7% of Accuracy.

Drive-by download is a sort of network attack which uses different techniques to plant malicious codes in their computers. It makes the traditional intrusion detection systems and firewalls nonfunctional in the reason that those devices could not detect web-based threats. The Crawler-based approach has been proposed by many studies to discover drive-by download sites. However, the Crawler-based approach could not simulate the real user behavior of web browsing when drive-by download attack happens. Therefore, this study proposes a new approach to detect drive-by download by sniffing HTTP flow. This study uses reputation system to improve the efficiency of client honeypots, and adjusts client honeypots to process the raw data of HTTP flow. In the experiment conducted in real network environment, this study show the performance of a single client honeypot could reach average 560,000 HTTP success access log per day. Even in the peak traffic, this mechanism reduced the process time to 22 hours, and detected drive-by download sites that users were actually browsing. Reputation system in this study is applicable to varieties of domain names because it does not refer to online WHOIS database. It established classification model on machine learning in 12 features. The correct classification rate of the reputation system applied in this study is 90.9%. Compared with other Reputation System studies, this study not only extract features from DNS A-Type but also extract features from DNS NS-Type. The experiment results show the Error Rate of the new features from DNS NS-Type is only 19.03%.

none

Recent improvements in structural genomics efforts have greatly increased the number of hypothetical proteins in the Protein Data Bank. Several computational methodologies have been developed to determine the function of these proteins but none of these methods have been able to account successfully for the diversity in the sequence and structural conformations observed in proteins that have the same function. An additional complication is the flexibility in both the protein active site and the ligand. In this dissertation, novel approaches to deal with both the ligand flexibility and the diversity in stereochemistry have been proposed. The active site analysis problem is formalized as a classification problem in which, for a given test protein, the goal is to predict the class of ligand most likely to bind the active site based on its stereochemical nature and thereby define its function. Traditional methods that have adapted a similar methodology have struggled to account for the flexibility observed in large ligands. Therefore, I propose a novel fragment-based approach to dealing with larger ligands. The advantage of the fragment-based methodology is that considering the protein-ligand interactions in a piecewise manner does not affect the active site patterns, and it also provides for a way to account for the problems associated with flexible ligands. I also propose two feature-based methodologies to account for the diversity observed in sequences and structural conformations among proteins with the same function. The feature-based methodologies provide detailed descriptions of the active site stereochemistry and are capable of identifying stereochemical patterns within the active site despite the diversity. Finally, I propose a Markov Random Field approach to combine the individual ligand fragment classifications (based on the stereochemical descriptors) into a single multi-fragment ligand class. This probabilistic framework combines the information provided by stereochemical features with the information regarding geometric constraints between ligand fragments to make a final ligand class prediction. The feature-based fragment identification methodology had an accuracy of 84% across a diverse set of ligand fragments and the mrf analysis was able to succesfully combine the various ligand fragments (identified by feature-based analysis) into one final ligand based on statistical models of ligand fragment distances. This novel approach to protein active site analysis was additionally tested on 3 proteins with very low sequence and structural similarity to other proteins in the PDB (a challenge for traditional methods) and in each of these cases, this approach successfully identified the cognate ligand. This approach addresses the two main issues that affect the accuracy of current automated methodologies in protein function assignment.

This paper presents a model transformation between System Dynamics Model (SDM) and Artificial Neural Network (ANN) to aid model construction and policy design. We first point out a similarity between a System Dynamics Model (SDM) and an artificial neural network, in which both store knowledge majorly in the structure (or linkages) of a model. Then, we design a method that can map a SDM to a special design Partial Recurrent Network (PRN), and prove in mathematics that they two operate under the same numerical propagation constraints. With the established foundation, we then showed that the SDM-PRN transformation could aid SDM construction in the following way: (1) start from an initial skeleton of a PRN model (mapping from an initial SDM), (2) incarnate its structure by learning and (3) convert it back to a corresponding SDM. This approach integrates the capability of neural network learning with a traditional process, which thus makes model construction more systematic and much easier for common people. In the same philosophy, the SDM-PRN transformation could also aid SD policy design. Since any PRN can learn some structures from a historical time series pattern, it can also learn a better structure from a better pattern set by designer. We have investigated the effectiveness and usefulness of two application of the SDM-PRN transformation described above and the results are satisfactory.

Computer-based training systems have become a mainstay in military and private institutions for training people how to perform certain complex tasks. As these tasks expand in difficulty, intelligent agents will appear as virtual teammates or tutors assisting a trainee in performing and learning the task. For developing these agents, we must obtain the strategies from expert players and emulate their behavior within the agent. Past researchers have shown the challenges in acquiring this information from expert human players and translating it into the agent. A solution for this problem involves using computer systems that assist in the human expert knowledge elicitation process. In this thesis, we present an approach for developing an agent for the game Revised Space Fortress, a game representative of the complex tasks found in training systems. Using machine learning techniques, the agent learns the strategy for the game by observing how a human expert plays. We highlight the challenges encountered while designing and training the agent in this real-time game environment, and our solutions toward handling these problems. Afterward, we discuss our experiment that examines whether trainees experience a difference in performance when training with a human or virtual partner, and how expert agents that express distinctive behaviors affect the learning of a human trainee. We show from our results that a partner agent that learns its strategy from an expert player serves the same benefit as a training partner compared to a programmed expert-level agent and a human partner of equal intelligence to the trainee.

Ribonucleic acid (RNA) sometimes occurs in a complex structure called pseudoknots. Prediction of RNA secondary structures has drawn much attention from both biologists and computer scientists. Consequently, many useful tools have been developed for RNA secondary structure prediction, with or without pseudoknots. These tools have their individual strength and weakness. As a result, we propose a hybrid feature extraction method which integrates two prediction tools pknotsRG and NUPACK with a support vector machine (SVM). We first extract some useful features from the target RNA sequence, and then decide its prediction tool preference with SVM classification. Our test data set contains 723 RNA sequences, where 202 pseudoknotted RNA sequences are obtained from PseudoBase, and 521 nested RNA sequences are obtained from RNA SSTRAND. Experimental results show that our method improves not only the overall accuracy but also the sensitivity and the selectivity of the target sequences. Our method serves as a preprocessing process in analyzing RNA sequences before employing the RNA secondary structure prediction tools. The ability to combine the existing methods and make the prediction tools more accurate is our main contribution.

In this document, we propose and examine the novel use of a learning mechanism between the reactive and deliberative layers of a hybrid robot control architecture. Balancing the need to achieve complex goals and meet real-time constraints, many modern mobile robot navigation control systems make use of a hybrid deliberative-reactive architecture. In this paradigm, a high-level deliberative layer plans routes or actions toward a known goal, based on accumulated world knowledge. A low-level reactive layer selects motor commands based on current sensor data and the deliberative layer's plan. The desired system-level effect of this architecture is that the robot is able to combine complex reasoning toward global objectives with quick reaction to local constraints. Implicit in this type of architecture, is the assumption that both layers are using the same model of the robot's capabilities and constraints. It may happen, for example, due to differences in representation of the robot's kinematic constraints, that the deliberative layer creates a plan that the reactive layer cannot follow. This sort of conflict may cause a degradation in system-level performance, if not complete navigational deadlock. Traditionally, it has been the task of the robot designer to ensure that the layers operate in a compatible manner. However, this is a complex, empirical task. Working to improve system-level performance and navigational robustness, we propose introducing a learning mechanism between the reactive layer and the deliberative layer, allowing the deliberative layer to learn a model of the reactive layer's execution of its plans. First, we focus on detecting this inter-layer conflict, and acting based on a corrected model. This is demonstrated on a physical robotic platform in an unstructured outdoor environment. Next, we focus on learning a model to predict instances of inter-layer conflict, and planning to act with respect to this model. This is demonstrated using supervised learning in a physics-based simulation environment. Results and algorithms are presented.

<p>This licentiate thesis presents a software architecture for inductive labeled dependency parsing of unrestricted natural language text, which achieves a strict modularization of parsing algorithm, feature model and learning method such that these parameters can be varied independently. The architecture is based on the theoretical framework of inductive dependency parsing by Nivre \citeyear{nivre06c} and has been realized in MaltParser, a system that supports several parsing algorithms and learning methods, for which complex feature models can be defined in a special description language. Special attention is given in this thesis to learning methods based on support vector machines (SVM).</p><p>The implementation is validated in three sets of experiments using data from three languages (Chinese, English and Swedish). First, we check if the implementation realizes the underlying architecture. The experiments show that the MaltParser system outperforms the baseline and satisfies the basic constraints of well-formedness. Furthermore, the experiments show that it is possible to vary parsing algorithm, feature model and learning method independently. Secondly, we focus on the special properties of the SVM interface. It is possible to reduce the learning and parsing time without sacrificing accuracy by dividing the training data into smaller sets, according to the part-of-speech of the next token in the current parser configuration. Thirdly, the last set of experiments present a broad empirical study that compares SVM to memory-based learning (MBL) with five different feature models, where all combinations have gone through parameter optimization for both learning methods. The study shows that SVM outperforms MBL for more complex and lexicalized feature models with respect to parsing accuracy. There are also indications that SVM, with a splitting strategy, can achieve faster parsing than MBL. The parsing accuracy achieved is the highest reported for the Swedish data set and very close to the state of the art for Chinese and English.</p> / <p>Denna licentiatavhandling presenterar en mjukvaruarkitektur för</p><p>datadriven dependensparsning, dvs. för att automatiskt skapa en</p><p>syntaktisk analys i form av dependensgrafer för meningar i texter</p><p>på naturligt språk. Arkitekturen bygger på idén att man ska kunna variera parsningsalgoritm, särdragsmodell och inlärningsmetod oberoende av varandra. Till grund för denna arkitektur har vi använt det teoretiska ramverket för induktiv dependensparsning presenterat av Nivre \citeyear{nivre06c}. Arkitekturen har realiserats i programvaran MaltParser, där det är möjligt att definiera komplexa särdragsmodeller i ett speciellt beskrivningsspråk. I denna avhandling kommer vi att lägga extra tyngd vid att beskriva hur vi har integrerat inlärningsmetoden supportvektor-maskiner (SVM).</p><p>MaltParser valideras med tre experimentserier, där data från tre språk används (kinesiska, engelska och svenska). I den första experimentserien kontrolleras om implementationen realiserar den underliggande arkitekturen. Experimenten visar att MaltParser utklassar en trivial metod för dependensparsning (\emph{eng}. baseline) och de grundläggande kraven på välformade dependensgrafer uppfylls. Dessutom visar experimenten att det är möjligt att variera parsningsalgoritm, särdragsmodell och inlärningsmetod oberoende av varandra. Den andra experimentserien fokuserar på de speciella egenskaperna för SVM-gränssnittet. Experimenten visar att det är möjligt att reducera inlärnings- och parsningstiden utan att förlora i parsningskorrekthet genom att dela upp träningsdata enligt ordklasstaggen för nästa ord i nuvarande parsningskonfiguration. Den tredje och sista experimentserien presenterar en empirisk undersökning som jämför SVM med minnesbaserad inlärning (MBL). Studien använder sig av fem särdragsmodeller, där alla kombinationer av språk, inlärningsmetod och särdragsmodell</p><p>har genomgått omfattande parameteroptimering. Experimenten visar att SVM överträffar MBL för mer komplexa och lexikaliserade särdragsmodeller med avseende på parsningskorrekthet. Det finns även vissa indikationer på att SVM, med en uppdelningsstrategi, kan parsa en text snabbare än MBL. För svenska kan vi rapportera den högsta parsningskorrektheten hittills och för kinesiska och engelska är resultaten nära de bästa som har rapporterats.</p>

Thesis (Ph.D.)--Utrecht University, 2006. / Includes bibliographical references (p. [133]-137).