Search

In this paper we introduce an efficient application for malicious PDF detection: ADEPT. With targeted attacks rising over the recent past, exploring a new detection and mitigation paradigm becomes mandatory. The use of malicious PDF files that exploit vulnerabilities in well-known PDF readers has become a popular vector for targeted attacks, for which few efficient approaches exist. Although simple in theory, parsing followed by analysis of such files is resource-intensive and may even be impossible due to several obfuscation and reader-specific artifacts. Our paper describes a new approach for detecting such malicious payloads that leverages machine learning techniques and an efficient feature selection mechanism for rapidly detecting anomalies. We assess our approach on a large selection of malicious files and report the experimental performance results for the developed prototype. [ABSTRACT FROM AUTHOR]

The management of atmospheric polluting reject is based on the ability to measure this pollution. This paper deals with the case where no local sensor can be used, inducing the use of video to detect and evaluate the atmospheric pollution coming from large industrial facilities. This paper presents a comparison of different classifiers used in a monitoring system of polluting smokes detected by cameras. The data used in this work are stemming from a system of video analysis and signal processing. The database also includes the pollution level of plumes of smoke defined by an expert. Several Machine Learning techniques are tested and compared. The experimental results are obtained from a real world database of polluting rejects. The parameters of each type of classifier are split in three categories: learned parameters, parameters determined by a first step of the experimentation, and parameters set by the programmer. The comparison of the results of the best classifier of each type indicates that all of them provide good results. [ABSTRACT FROM AUTHOR]

In this paper the fusion of artificial neural networks, granular computing and learning automata theory is proposed and we present as a final result ANLAGIS, an adaptive neuron-like network based on learning automata and granular inference systems. ANLAGIS can be applied to both pattern recognition and learning control problems. Another interesting contribution of this paper is the distinction between pre-synaptic and post-synaptic learning in artificial neural networks. To illustrate the capabilities of ANLAGIS some experiments on knowledge discovery in data mining and machine learning are presented. Previous work of Jang et al. [1] on adaptive network-based fuzzy inference systems, or simply ANFIS, can be considered a precursor of ANLAGIS. The main, novel contribution of ANLAGIS is the incorporation of Learning Automata Theory within its structure. [ABSTRACT FROM AUTHOR]

This paper focuses on classification tasks. The goal of the paper is to propose a framework for adaptive and integrated machine classification and to investigate the effect of different adaptation and integration schemes. After having introduced several integration and adaptation schemes a framework for adaptive and integrated classification in the form of the software shell is proposed. The shell allows for integrating data pre-processing with data mining stages using population-based and A-Team techniques. The approach was validated experimentally. Experiment results have shown that integrated and adaptive classification outperforms traditional approaches. [ABSTRACT FROM AUTHOR]

The purpose of this paper is to present a methodology by which rotating machinery faults can be diagnosed. The proposed method is based on random forests algorithm (RF), a novel assemble classifier which builds a large amount of decision trees to improve on the single tree classifier. Although there are several existed techniques for faults diagnosis, such as artificial neural network, support vector machines etc, the research on RF is meaningful and necessary because of its fast executed speed, the characteristic of tree classifier, and high performance in machine faults diagnosis. Evaluation of the RF based method has been demonstrated by a case study on induction motors faults diagnosis. Experimental results indicate the validity and reliability of RF based fault diagnosis methodology. Furthermore, an optimized form of RF is also provided in this paper. We employ the genetic algorithm to strengthen RF. The capability of optimized RF algorithm is proved with same experimental data. It is the evidence that RF based diagnosis methodology can generate more accurate outcome by combining with other optimization method. [ABSTRACT FROM AUTHOR]

Multilabel classification is an important problem in bioinformatics and Machine Learning. In a conventional classification problem, examples belong to just one among many classes. When an example can simultaneously belong to more than one class, the classification problem is named multilabel classification problem. Protein function classification is a typical example of multilabel classification, since a protein may have more than one function. This paper describes the main characteristics of some multilabel classification methods and applies five methods to protein classification problems. For an experimental comparison of these methods, traditional machine learning techniques are used. The paper also compares different evaluation metrics used in multilabel problems. [ABSTRACT FROM AUTHOR]

The development of effective knowledge adaption techniques is one of the promising solutions to improve the performance of case-based reasoning (CBR) systems. Case-base maintenance becomes a powerful method to refine knowledge in CBR systems. This paper proposes an integrated knowledge adaption framework for CBR systems which contains a meta database component and a maintenance strategies component. The meta database component can help track changes of interested concepts and therefore enable a CBR system to signal a need for maintenance or to invoke adaption on its own. The maintenance strategies component can perform cross-container maintenance operations in a CBR system. This paper also illustrates how the proposed integrated knowledge adaption framework assists decision makers to build dynamic prediction and decision capabilities. [ABSTRACT FROM AUTHOR]

In Morpho Challenge competitions, the objective has been to design statistical machine learning algorithms that discover which morphemes (smallest individually meaningful units of language) words consist of. Ideally, these are basic vocabulary units suitable for different tasks, such as text understanding, machine translation, information retrieval (IR), and statistical language modeling. In this paper, we propose to evaluate the morpheme analyses by performing IR experiments, where the words in the documents and queries are replaced by their proposed morpheme representations and the search is based on morphemes instead of words. In this paper, the evaluations are run for three languages: Finnish, German, and English using the queries, texts, and relevance judgments available in CLEF forum. The results show that the morpheme analysis has a significant effect in IR performance in all languages, and that the performance of the best unsupervised methods can be superior to the supervised reference methods. [ABSTRACT FROM AUTHOR]

In order to classify an unseen (query) vector q with the k-Nearest Neighbors method (k-NN) one computes a similarity function between q and training vectors in a database. In the basic variant of the k-NN algorithm the predicted class of q is estimated by taking the majority class of the q΄s k-nearest neighbors. Various similarity functions may be applied leading to different classification results. In this paper a heterogeneous similarity function is constructed out of different 1-component metrics by minimization of the number of classification errors the system makes on a training set. The HSFL-NN system, which has been introduced in this paper, on five tested datasets has given better results on unseen samples than the plain k-NN method with the optimally selected k parameter and the optimal homogeneous similarity function. [ABSTRACT FROM AUTHOR]

The software agents are applied for a remote search of information. It seems natural that to analyse such information machine learning routines should be built-in into an agent system. After finding and processing the data the generated rules will be evaluated by means of so called interestingness measures, and only the best rules should be returned to the user. The paper presents situation in civil engineering data processing, as a suggestion for designers of intelligent software tools, to work out difficult but much needed procedures that should be implemented into autonomous agent system, intended for retrieving special kind of information searched for example by materials technologists. A simple architecture for an agent system is suggested without, however, getting into any technical details on how the elements of such system should be constructed. [ABSTRACT FROM AUTHOR]

Uranium extraction in Kazakhstan is carried out using underground leaching method. Economic performance depends on the production process speed and accuracy of geophysical data interpretation. Data interpretation can be performed using learned systems, such as artificial neural network (ANN), Linear Discriminant Analysis Classifier (LDAC), Support Vector Classification (SVM), k-Nearest-Neighbor (k-NN) and etc. In the paper ˵adjacency cube″ method for integration of results of few interpretation algorithms is proposed. Learning algorithm for the ˵adjacency cube″ with low computational complexity was developed. The proposed method improves quality of recognition by 2-3 percent. [ABSTRACT FROM AUTHOR]

Explanatory analogies make learning complex concepts easier by elaborately mapping a target concept onto a more familiar source concept. Solutions exist for automatically retrieving shorter metaphors from natural language text, but not for explanatory analogies. In this paper, we propose an approach to find webpages containing explanatory analogies for a given target concept. For this, we propose the use of a `region of interest' (ROI) based on the observation that linguistic markers and source concept often co-occur with various forms of the word `analogy'. We also suggest an approach to identify the source concept(s) contained in a retrieved analogy webpage. We demonstrate these approaches on a dataset created using Google custom search to find candidate web pages that may contain analogies. [ABSTRACT FROM AUTHOR]

In almost all workflow management system the role concept is determined once at the introduction of workflow application and is not reevaluated to observe how successfully certain processes are performed by the authorized agents. This paper describes an approach which evaluates how agents are working successfully and feed this information back for future agent assignment to achieve maximum business benefit for the enterprise. The approach is called Pattern based Agent Performance Evaluation (PAPE) and is based on machine learning technique combined with post processing technique. We report on the result of our experiments and discuss issues and improvement of our approach. [ABSTRACT FROM AUTHOR]

This paper presents a knowledge extraction system for providing sales intelligence based on information downloaded from the WWW. The information is first located and downloaded from relevant companies΄ websites and then machine learning is used to find these web pages that contain useful information where useful is defined as containing news about orders for specific products. Several machine learning algorithms were tested from which k-nearest neighbour, support vector machines, multi-layer perceptron and C4.5 decision tree produced best results in one or both experiments however k-nearest neighbour and support vector machines proved to be most robust which is a highly desired characteristic in the particular application. K-nearest neighbour slightly outperformed the support vector machines in both experiments which contradicts the results reported previously in the literature. [ABSTRACT FROM AUTHOR]

This paper outlines research in progress intended to contribute to the autonomous management of networks, allowing policies to be dynamically adjusted and aligned to application directives according to the available resources. Many existing management approaches require static a priori policy deployment but our proposal goes one step further modifying initially deployed policies by learning from the system behaviour. We use a hierarchical policy model to show the connection of high level goals with network level configurations. We also intend to solve two important and mostly forgotten issues: the system has multiple goals some of them contradictory and we will show how to overcome it; and, some current works optimize one network element but being unaware of other participants; instead, our proposed scheme takes into account various social behaviours, such as cooperation and competition among different elements. [ABSTRACT FROM AUTHOR]

We have developed some projects [1,2] for addressing the heterogeneity problem in open learning management systems (LMS). In [3], an independent adaptation platform to support competences development through personalization is presented. Three user characteristics (competences profile, learning style, and accessing context) are modeled by means of analyzing user interaction data in a LMS. This process is supported by the assigment of independent adaptation tasks to different JADE intelligent agents. In this paper we introduce some negotiation strategies among those intelligent agents in order to: 1) select the best types of adaptation through collaborative tasks, and 2) generate standards and exchangeable user profiles based on the inferred user characteristics, describing the mechanisms to mobilize these profiles between different LMSs. These profiles support the generation of specifics learning designs for each particular user. [ABSTRACT FROM AUTHOR]

Cyclone is a mixed-initiative and adaptive clustering and structure generation environment which is capable of learning categorization behavior through user interaction as well as conducting auto-categorization based on the extracted model. The strength of Cyclone resides in its integration of several visualization and interface techniques with data mining and AI learning processes. This paper presents the intuitive visual interface of Cyclone which empowers the user to explore, analyze, exploit and structure unstructured information from various sources generating a personalized taxonomy in real-time and on-the-fly. [ABSTRACT FROM AUTHOR]

In this paper, we present an adaptation of Gaussian Processes for learning a joint probabilistic distribution using Bayesian Programming. More specifically, a robot navigation problem will be showed as a case of study. In addition, Gaussian Processes will be compared with one of the most popular techniques for machine learning: Neural Networks. Finally, we will discuss about the accuracy of these methods and will conclude proposing some future lines for this research. [ABSTRACT FROM AUTHOR]

In this paper, a new co-training style semi-supervised algorithm is proposed, which employs Bagging based multimodal perturbation to label the unlabeled data. In detail, through perturbing the training data, input attributes and learning parameters together, the algorithm generates accurate but diversity k-nearest neighbor classifiers. These classifiers are refined using unlabeled examples which are labeled if the other classifiers agree on the labeling. Experimental results show that the semi-supervised algorithm could effectively improve the classification generalization by utilizing the unlabeled data. [ABSTRACT FROM AUTHOR]

This paper presents a corpus-based approach for automatic unknown word recognition in Thai. This approach applies an ensemble learning technique to generate a model for classifying unknown word candidates using features obtained from a corpus. We propose a technique called ˵group-based evaluation by ranking″. It clusters the unknown word candidates into groups based on the occuring locations. The candidate with the highest accuracy is then identified as an unknown word. In this task, the number of positive instances is dominantly smaller than that of negative instances, forming an unbalanced data set. To improve the prediction accuracy, we apply a boosting technique with ˵voting under group-based evaluation by ranking″. We have conducted experiments on real-world data to evaluate the performance of the proposed approach. The experiments compared the accuracy of our technique with an ordinary naïve Bayes technique. Our technique achieves the accuracy 90.93±0.50% when the first rank is selected and 97.90±0.26% when the candidates up to the tenth rank are considered. This is 6.79% to 8.45% improvement. [ABSTRACT FROM AUTHOR]

Since the amount of information is rapidly growing, there is an overwhelming interest in efficient network computing systems including Grids, public-resource computing systems, P2P systems and cloud computing. In this paper we take a detailed look at the problem of modeling and optimization of network computing systems for parallel decision tree induction methods. Firstly, we present a comprehensive discussion on mentioned induction methods with a special focus on their parallel versions. Next, we propose a generic optimization model of a network computing system that can be used for distributed implementation of parallel decision trees. To illustrate our work we provide results of numerical experiments showing that the distributed approach enables significant improvement of the system throughput. [ABSTRACT FROM AUTHOR]

Recently, building sparse SVMs becomes an active research topic due to its potential applications in large scale data mining tasks. One of the most popular approaches to building sparse SVMs is to select a small subset of training samples and employ them as the support vectors. In this paper, we explain that selecting the support vectors is equivalent to selecting a number of columns from the kernel matrix, and is equivalent to selecting a subset of features in the feature selection domain. Hence, we propose to use an effective feature selection algorithm, namely the Minimum Redundancy – Maximum Relevance (MRMR) algorithm to solve the support vector selection problem. MRMR algorithm was then compared to two existing methods, namely back-fitting (BF) and pre-fitting (PF) algorithms. Preliminary results showed that MRMR generally outperformed BF algorithm while it was inferior to PF algorithm, in terms of generalization performance. However, the MRMR approach was extremely efficient and significantly faster than the two compared algorithms. [ABSTRACT FROM AUTHOR]

Recursive Neural Networks are non-linear adaptive models that are able to learn deep structured information. However, these models have not yet been broadly accepted. This fact is mainly due to its inherent complexity. In particular, not only for being extremely complex information processing models, but also because of a computational expensive learning phase. The most popular training method for these models is back-propagation through the structure. This algorithm has been revealed not to be the most appropriate for structured processing due to problems of convergence, while more sophisticated training methods enhance the speed of convergence at the expense of increasing significantly the computational cost. In this paper, we firstly perform an analysis of the underlying principles behind these models aimed at understanding their computational power. Secondly, we propose an approximate second order stochastic learning algorithm. The proposed algorithm dynamically adapts the learning rate throughout the training phase of the network without incurring excessively expensive computational effort. The algorithm operates in both on-line and batch modes. Furthermore, the resulting learning scheme is robust against the vanishing gradients problem. The advantages of the proposed algorithm are demonstrated with a real-world application example. [ABSTRACT FROM AUTHOR]

Most machine learning algorithms are essentially generalized learning schemes. When used incrementally, conflict between the concepts already learned and the new events to be learned need be eliminated. This paper presents a new incremental learning scheme that divides induction into two phases: incrementally accurate learning and generalization. The new learning scheme is conflict-free thus it is suitable for incremental induction. [ABSTRACT FROM AUTHOR]

This paper is concerned with the fuzzy support vector classification, in which both of the type of the output training point and the value of the final fuzzy classification function are triangle fuzzy number. First, the fuzzy classification problem is formulated as a fuzzy chance constrained programming. Then, we transform this programming into its equivalence quadratic programming. Final, a fuzzy support vector classification algorithm is proposed to deal with the problem. An example is presented to illustrate rationality of the algorithm. [ABSTRACT FROM AUTHOR]

Thousands of Machine Learning research papers contain experimental comparisons that usually have been conducted with a single focus of interest, often losing detailed results after publication. Yet, when collecting all these past experiments in experiment databases, they can readily be reused for additional and possibly much broader investigation. In this paper, we make use of such a database to answer various interesting research questions about learning algorithms and to verify a number of recent studies. Alongside performing elaborate comparisons of algorithms, we also investigate the effects of algorithm parameters and data properties, and seek deeper insights into the behavior of learning algorithms by studying their learning curves and bias-variance profiles. [ABSTRACT FROM AUTHOR]

As network attacks have increased in number and severity over the past few years, intrusion detection system (IDS) is increasingly becoming a critical component to secure the network. Due to large volumes of security audit data as well as complex and dynamic properties of intrusion behaviors, optimizing performance of IDS becomes an important open problem that is receiving more and more attention from the research community. The uncertainty to explore if certain algorithms perform better for certain attack classes constitutes the motivation for the reported herein. In this paper, we evaluate performance of a comprehensive set of classifier algorithms using KDD99 dataset. Based on evaluation results, best algorithms for each attack category is chosen and two classifier algorithm selection models are proposed. The simulation result comparison indicates that noticeable performance improvement and real-time intrusion detection can be achieved as we apply the proposed models to detect different kinds of network attacks. [ABSTRACT FROM AUTHOR]

One key problem for incremental learning is to maintain concept consistency. AQ learning algorithm based on star is essentially a generalized learning scheme. When used incrementally, it needs to eliminate conflict between the concept already learned and the new events to be learnt. This paper presents a new learning scheme that learns non-expandable DNF incrementally and is of conflict-free. [ABSTRACT FROM AUTHOR]

One of the definitely unsolved main problems in molecular biology is the protein-protein functional association prediction problem. Genetic Programming (GP) is applied to this domain. GP evolves an expression, equivalent to a binary classifier, which predicts if a given pair of proteins interacts. We take advantages of GP flexibility, particularly, the possibility of defining new operations. In this paper, the missing values problem benefits from the definition of if-unknown, a new operation which is more appropriate to the domain data semantics. Besides, in order to improve the solution size and the computational time, we use the Tarpeian method which controls the bloat effect of GP. According to the obtained results, we have verified the feasibility of using GP in this domain, and the enhancement in the search efficiency and interpretability of solutions due to the Tarpeian method. [ABSTRACT FROM AUTHOR]

On-line behavioural targeting is a dynamically evolving area of web mining concerning the applications of data analysis of on-line users΄ behaviour and machine learning in optimising web on-line advertising and constitutes a problem of high importance and complexity. The paper reports on experimental work concerning testing various benchmark machine-learning algorithms and attribute preprocessing techniques in the context of behavioural targeting. Our final goal is to build a system which automatically learns and subsequently decides which on-line advertisements to present to a user visiting a web page, based on his previous recorded behaviour, in order to maximise the revenue of the ad-network and the web site hosting the ad, and, at the same time, to minimise the user΄s annoyance caused by potentially inappropriate advertisements. We present a general adaptive model which makes it possible to test various machine-learning algorithms in a plug-in mode and its implementation. We also report our experimental work concerning comparison of the performance of various machine-learning algorithms and data preprocessing techniques on a real dataset. The performance of our experiments is evaluated by some objective metrics such as the click-through rate (CTR) or related. Our experimental results clearly indicate that the presented adaptive system can significantly increase the CTR metric by 40% for some settings. All the experiments are performed on a real industrial dataset concerning on-line ads emitted in the Polish Web during a 1-month period in 2007. Up to the authors΄ best knowledge this is the first and largest evaluation made on this kind of real industrial data in Poland, at the time of writing. [ABSTRACT FROM AUTHOR]

In this paper, we propose a machine learning method to knowledge element extraction from learning resources. First, we build a knowledge element taxonomy containing 25 semantic types. Second, we formalize the knowledge element extraction of single semantic type as binary classification. Finally, we construct the multi-class classification model which can predict the semantic type of knowledge element by merge the results of binary classifiers. We annotate three semantic types in corpus and use them as training data, train the machine learning models. In experiment, we compared three binary classification models: Decision Tree, SVM and Naïve Bayesian. The experimental results show that SVM has better average performance. We employ ECOC method to construct multi-class classification model and use SVM as base binary classifier in the model. Our approach outperforms the baseline in experiment. The experimental results indicate that our approach is effective. [ABSTRACT FROM AUTHOR]

This paper presents a real-time news event extraction system developed by the Joint Research Centre of the European Commission. It is capable of accurately and efficiently extracting violent and disaster events from online news without using much linguistic sophistication. In particular, in our linguistically relatively lightweight approach to event extraction, clustered news have been heavily exploited at various stages of processing. The paper describes the system΄s architecture, news geo-tagging, automatic pattern learning, pattern specification language, information aggregation, the issues of integrating event information in a global crisis monitoring system and new experimental evaluation. [ABSTRACT FROM AUTHOR]

In this paper, a new approach for multi-class classification problems is applied to the Semantic Role Labeling (SRL) problem, which is an important task for natural language processing systems to achieve better semantic understanding of text. The new approach applies to any classification problem with large feature sets. Data is partitioned using clusters on a subset of the features. A multi-label classifier is then trained individually on each cluster, using automatic feature selection to customize the larger feature set for the cluster. This algorithm is applied to the Semantic Role Labeling problem and achieves improvements in accuracy for both the argument identification classifier and the argument labeling classifier. [ABSTRACT FROM AUTHOR]

Potential distribution modelling has been widely used to predict and to understand the geographical distribution of species. These models are generally produced by retrieving the environmental conditions where the species is known to be present or absent and feeding this data into a modelling algorithm. This paper investigates the use of Machine Learning techniques in the potential distribution modelling of plant species Stryphnodendron obovatum Benth (MIMOSACEAE). Three techniques were used: Support Vector Machines, Genetic Algorithms and Decision Trees. Each technique was able to extract a different representation of the relations between the environmental conditions and the distribution profile of the species being considered. [ABSTRACT FROM AUTHOR]

Many data mining techniques are these days in use for ontology learning – text mining, Web mining, graph mining, link analysis, relational data mining, and so on. In the current state-of-the-art bundle there is a lack of ˵software mining″ techniques. This term denotes the process of extracting knowledge out of source code. In this paper we approach the software mining task with a combination of text mining and link analysis techniques. We discuss how each instance (i.e. a programming construct such as a class or a method) can be converted into a feature vector that combines the information about how the instance is interlinked with other instances, and the information about its (textual) content. The so-obtained feature vectors serve as the basis for the construction of the domain ontology with OntoGen, an existing system for semi-automatic data-driven ontology construction. [ABSTRACT FROM AUTHOR]

All learning algorithms perform very well when provided with a small number of highly relevant features. This paper proposes a constructive induction method to automatically construct such features. The method, named GLOREF (GLObally RElevant Features), exploits low-level interactions between the attributes in order to generate globally relevant features. The usefulness of the approach is demonstrated empirically through a large scale experiment involving 13 classifiers and 24 datasets. Results demonstrate the ability of the method in generating highly informative features and a strong positive effect on the accuracy of the classifiers. [ABSTRACT FROM AUTHOR]

This paper discusses the problem of predicting the structural changes in an ontology. It addresses ontologies that contain instances in addition to concepts. The focus is on an ontology where the instances are textual documents, but the approach presented in this document is general enough to also work with other kinds of instances, as long as a similarity measure can be defined over them. We examine the changes in the Open Directory Project ontology of Web pages over a period of several years and analyze the most common types of structural changes that took place during that time. We then present an approach for predicting one of the more common types of structural changes, namely the addition of a new concept that becomes the subconcept of an existing parent concept and adopts a few instances of this existing parent concept. We describe how this task can be formulated as a machine-learning problem and present an experimental evaluation of this approach that shows promising results of the proposed approach. [ABSTRACT FROM AUTHOR]

Random Forests, Support Vector Machines and k-Nearest Neighbors are successful and proven classification techniques that are widely used for different kinds of classification problems. One of them is classification of genomic and proteomic data that is known as a problem with extremely high dimensionality and therefore demands suited classification techniques. In this domain they are usually combined with gene selection techniques to provide optimal classification accuracy rates. Another reason for reducing the dimensionality of such datasets is their interpretability. It is much easier to interpret a small set of ranked genes than 20 or 30 thousands of unordered genes. In this paper we present a classification ensemble of decision trees called Rotation Forest and evaluate its classification performance on small subsets of ranked genes for 14 genomic and proteomic classification problems. An important feature of Rotation Forest is demonstrated – i.e. robustness and high classification accuracy using small sets of genes. [ABSTRACT FROM AUTHOR]

With the growing use of computers and the Internet, it has become difficult for organizations to locate and effectively manage sensitive personally identifiable information (PII). This problem becomes even more evident in collaborative computing environments. PII may be hidden anywhere within the file system of a computer. As well, in the course of different activities, via collaboration or not, personally identifiable information may migrate from computer to computer. This makes meeting the organizational privacy requirements all the more complex. Our particular interest is to develop technology that would automatically discover workflow across organizational collaborators that would include private data. Since in this context, it is important to understand where and when the private data is discovered, in this paper, we focus on PII discovery, i.e. automatically identifying private data existant in semi-structured and unstructured (free text) documents. The first part of the process involves identifying PII via named entity recognition. The second part determines relationships between those entities based upon a supervised machine learning method. We present test results of our methods using publicly-available data generated from different collaborative activities to provide an assessment of scalability in cooperative computing environment. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of supervised sentiment detection using classifiers which are derived from word features. We argue that, while the literature has suggested the use of lexical features is inappropriate for sentiment detection, a careful and thorough evaluation reveals a less clear-cut state of affairs. We present results from five classifiers using word-based features on three tasks, and show that the variation between classifiers can often be as great as has been reported between different feature sets with a fixed classifier. We are thus led to conclude that classifier choice plays at least as important a role as feature choice, and that in many cases word-based classifiers perform well on the sentiment detection task. [ABSTRACT FROM AUTHOR]

Hypercountors are methods for automatic image analysis. They can be interpreted as contextual classifiers that use expert knowledge and operate in supervised or unsupervised mode. In the present paper, it is shown that hypercontours may act as agents cooperating in recognition of objects placed on computer images. [ABSTRACT FROM AUTHOR]

In this paper, we would like to introduce a multi-agent mechanism to protect target systems from computer virus infections. Using a machine learning approach, we first define a form of object knowledge to determine computer viruses and diagnosed objects. Second, we set up an association model of knowledge base and database. The database stores information of diagnosed objects. The knowledge base contains certain sets of deduction rules. Finally, we build two active agents to control virus infections. In an event-learning model, the first agent named Virus Auto-protect Agent is used to monitor all suspicious events. The second one, Virus Scanning Agent is used in an explanation-learning model to scan for viruses, to warn users of dangers and to restore the data from the previous state of safety. The experimentation results show that the anti-virus system can quickly recognize known and unknown computer virus infections. [ABSTRACT FROM AUTHOR]

Spam filtering is defined as a task trying to label emails with spam or ham in an online situation. The online feature requires the spam filter has a strong timely generalization and has a high processing speed. Machine learning can be employed to fulfill the two requirements. In this paper, we propose a SVMEL (SVM Ensemble Learning) method to combine five simple filters for higher accuracy and an active learning method to choose training emails for less training time. The experiments results show the filter applying active learning method can reduce requirements of labeled training emails and reach steady-state performance more quickly. [ABSTRACT FROM AUTHOR]

Today΄s tendency to protect various copyrighted multimedia contents, such as text, images or video, resulted in many algorithms for detecting duplicates. If the observed content is identical, then the task is easy. But if the content is even slightly changed, the task to identify the duplicate can be difficult and time consuming. In this paper we develop a fast, two-step algorithm for detecting image duplicates. The algorithm finds also slightly changed images with added noise, translated or scaled content, or images having been compressed and decompressed by various algorithms. The time needed to detect duplicates is kept low by implementing image feature-based searches. To detect all similar images for a given reference image, the feature extraction based on convex layers is deployed. The correlation coefficient between two features gives the first hint of similarity to the user, who creates a learning set for support vector machines by simple on-screen selection. [ABSTRACT FROM AUTHOR]

Knowledge acquisition is a critical problem for machine translation and translation selection. In this paper, I propose a tranlsation selection method that combines variable features from multiple language resources using machine learning. I introduce multiple measures for sense disambiguation and word selection that are based on language resources, and apply machine learning to combine those measures for translation selection. In evaluation, precision of translation selection improves even though a small-sized bilingual corpus is used as training data. [ABSTRACT FROM AUTHOR]

This paper shows how big data can be experimentally used at large scale for marketing purposes at a mobile network operator. We present results from a large-scale experiment in a MNO in Asia where we use machine learning to segment customers for text-based marketing. This leads to conversion rates far superior to the current best marketing practices within MNOs. Using metadata and social network analysis, we created new metrics to identify customers that are the most likely to convert into mobile internet users. These metrics falls into three categories: discretionary income, timing, and social learning. Using historical data, a machine learning prediction model is then trained, validated, and used to select a treatment group. Experimental results with 250 000 customers show a 13 times better conversion-rate compared to the control group. The control group is selected using the current best practice marketing. The model also shows very good properties in the longer term, as 98% of the converted customers in the treatment group renew their mobile internet packages after the campaign, compared to 37% in the control group. These results show that data-driven marketing can significantly improve conversion rates over current best-practice marketing strategies. [ABSTRACT FROM AUTHOR]

Remote desktop connection (RDC) services offer clients access to remote content and services, commonly used to access their working environment. With the advent of cloud-based services, an example use case is that of delivering virtual PCs to users in WAN environments. In this paper, we aim to analyze common user behavior when accessing RDC services. We first identify different behavioral categories, and conduct traffic analysis to determine a feature set to be used for classification purposes. We then propose a machine learning approach to be used for classifying behavior, and use this approach to classify a large number of real-world RDCs. Obtained results may be applied in the context of network resource planning, as well as in making Quality of Experience-driven resource allocation decisions. [ABSTRACT FROM AUTHOR]

Vision-based hand gesture recognition is an area of active current research in computer vision and machine learning. Being a natural way of human interaction, it is an area where many researchers are working on, with the goal of making human computer interaction (HCI) easier and natural, without the need for any extra devices. So, the primary goal of gesture recognition research is to create systems, which can identify specific human gestures and use them, for example, to convey information. For that, vision-based hand gesture interfaces require fast and extremely robust hand detection, and gesture recognition in real time. Hand gestures are a powerful human communication modality with lots of potential applications and in this context we have sign language recognition, the communication method of deaf people. Sign languages are not standard and universal and the grammars differ from country to country. In this paper, a real-time system able to interpret the Portuguese Sign Language is presented and described. Experiments showed that the system was able to reliably recognize the vowels in real-time, with an accuracy of 99.4% with one dataset of features and an accuracy of 99.6% with a second dataset of features. Although the implemented solution was only trained to recognize the vowels, it is easily extended to recognize the rest of the alphabet, being a solid foundation for the development of any vision-based sign language recognition user interface system. [ABSTRACT FROM AUTHOR]

Quality-of-experience (QoE) produces the blue print of human perception, feelings, needs and intentions, while Quality-of-Service (QoS) is a technology centric metric used to assess the performance of a multimedia services and/or network. .It is quite important for service/content providers to understand user/customer experience requirements in order to improve the service quality or the content recommendation. With advent of 3G and 4G wireless networks, and efficient smart phones, the band-width hungry multimedia applications are becoming common in use on end-user devices. Thus,it is also important for telecom operators to understand the impact of wireless network performances on the user experience in mobile environment. On the fly evaluation of user experience for multimedia services is a challenging problem especially in mobile environments. It implies the collection and the correlation of a mixture of variables on network conditions, on the service, as well as on the user itself. This paper proposes an innovative mobile application that can be used for measuring user quality-of-experience on the fly with a high accuracy and the consideration of multiple parameters about the user, the network and the system. This application takes advantages of current advances in mobile technologies to measure user experience directly on the user device. In addition, it aims to preserve the user privacy by transmitting only estimated quality-of-experience to the service provider. [ABSTRACT FROM AUTHOR]

As the landscape around Big data continues to exponentially evolve, the « big » facet of Big data is no more number one priority of researchers and IT professionals. The race has recently become more about how to sift through torrents of data to find the hidden diamond and engineer a better, smarter and healthier world. The ease with which our mobile captures daily data about ourselves makes it an exceptionally suitable means for ultimately improving the quality of our lives and gaining valuable insights into our affective, mental and physical state. This paper takes the first exploratory step into this direction by using the mobile to process and analyze the ˵digital exhaust″ it collects to automatically recognize our emotional states and accordingly respond to them in the most effective and ˵human″ way possible. To achieve this we treat all technical, psycho-somatic, and cognitive aspects of emotion observation and prediction, and repackage all these elements into a mobile multimodal emotion recognition system that can be used on any mobile device. [ABSTRACT FROM AUTHOR]