In this work we study two aspects of image security: improper usage and illegal access of images. In the first part we present our results on steganalysis – protection against improper usage of images. In the second part we present our results on image encryption – protection against illegal access of images. Steganography is the collective name for methodologies that allow the creation of invisible –hence secret– channels for information transfer. Steganalysis, the counter to steganography, is a collection of approaches that attempt to detect and quantify the presence of hidden messages in cover media. First we present our studies on stego-images using features developed for data stream classification towards making some qualitative assessments about the effect of steganography on the lower order bit planes(LSB) of images. These features are effective in classifying different data streams. Using these features, we study the randomness properties of image and stego-image LSB streams and observe that data stream analysis techniques are inadequate for steganalysis purposes. This provides motivation to arrive at steganalytic techniques that go beyond the LSB properties. We then present our steganalytic approach which takes into account such properties. In one such approach, we perform steganalysis from the point of view of quantifying the effect of perturbations caused by mild image processing operations–zoom-in/out, rotation, distortions–on stego-images. We show that this approach works both in detecting and estimating the presence of stego-contents for a particularly difficult steganographic technique known as LSB matching steganography. Next, we present our results on our image encryption techniques. Encryption approaches which are used in the context of text data are usually unsuited for the purposes of encrypting images(and multimedia objects) in general. The reasons are: unlike text, the volume to be encrypted could be huge for images and leads to increased computational requirements; encryption used for text renders images incompressible thereby resulting in poor use of bandwidth. These issues are overcome by designing image encryption approaches that obfuscate the image by intelligently re-ordering the pixels or encrypt only parts of a given image in attempts to render them imperceptible. The obfuscated image or the partially encrypted image is still amenable to compression. Efficient image encryption schemes ensure that the obfuscation is not compromised by the inherent correlations present in the image. Also they ensure that the unencrypted portions of the image do not provide information about the encrypted parts. In this work we present two approaches for efficient image encryption. First, we utilize the correlation preserving properties of the Hilbert space-filling-curves to reorder images in such a way that the image is obfuscated perceptually. This process does not compromise on the compressibility of the output image. We show experimentally that our approach leads to both perceptual security and perceptual encryption. We then show that the space-filling curve based approach also leads to more efficient partial encryption of images wherein only the salient parts of the image are encrypted thereby reducing the encryption load. In our second approach, we show that Singular Value Decomposition(SVD) of images is useful from the point of image encryption by way of mismatching the unitary matrices resulting from the decomposition of images. It is seen that the images that result due to the mismatching operations are perceptually secure.

Frequent-pattern mining from databases has been widely studied and frequently observed. Unfortunately, these algorithms are not suitable for data stream processing. In frequent-pattern mining from data streams, it is important to manage sets of items and also their history. There are several reasons for this; it is not just the history of frequent items, but also the history of potentially frequent sets that can become frequent later. This requires more memory and computational power. This thesis describes two algorithms: Lossy Counting and FP-stream. An effective implementation of these algorithms in C# is an integral part of this thesis. In addition, the two algorithms have been compared.

Dans cette thèse, nous nous intéressons aux méthodes dites récursives qui permettent une mise à jour des estimations séquentielles de données spatiales ou spatio-temporelles et qui ne nécessitent pas un stockage permanent de toutes les données. Traiter et analyser des flux des données, Data Stream, de façon effective et efficace constitue un défi actif en statistique. En effet, dans beaucoup de domaines d'applications, des décisions doivent être prises à un temps donné à la réception d'une certaine quantité de données et mises à jour une fois de nouvelles données disponibles à une autre date. Nous proposons et étudions ainsi des estimateurs à noyau de la fonction de densité de probabilité et la fonction de régression de flux de données spatiales ou spatio-temporelles. Plus précisément, nous adaptons les estimateurs à noyau classiques de Parzen-Rosenblatt et Nadaraya-Watson. Pour cela, nous combinons la méthodologie sur les estimateurs récursifs de la densité et de la régression et celle d'une distribution de nature spatiale ou spatio-temporelle. Nous donnons des applications et des études numériques des estimateurs proposés. La spécificité des méthodes étudiées réside sur le fait que les estimations prennent en compte la structure de dépendance spatiale des données considérées, ce qui est loin d'être trivial. Cette thèse s'inscrit donc dans le contexte de la statistique spatiale non-paramétrique et ses applications. Elle y apporte trois contributions principales qui reposent sur l'étude des estimateurs non-paramétriques récursifs dans un cadre spatial/spatio-temporel et s'articule autour des l'estimation récursive à noyau de la densité dans un cadre spatial, l'estimation récursive à noyau de la densité dans un cadre spatio-temporel, et l'estimation récursive à noyau de la régression dans un cadre spatial. / In this thesis, we are interested in recursive methods that allow to update sequentially estimates in a context of spatial or spatial-temporal data and that do not need a permanent storage of all data. Process and analyze Data Stream, effectively and effciently is an active challenge in statistics. In fact, in many areas, decisions should be taken at a given time at the reception of a certain amount of data and updated once new data are available at another date. We propose and study kernel estimators of the probability density function and the regression function of spatial or spatial-temporal data-stream. Specifically, we adapt the classical kernel estimators of Parzen-Rosenblatt and Nadaraya-Watson. For this, we combine the methodology of recursive estimators of density and regression and that of a distribution of spatial or spatio-temporal data. We provide applications and numerical studies of the proposed estimators. The specifcity of the methods studied resides in the fact that the estimates take into account the spatial dependence structure of the relevant data, which is far from trivial. This thesis is therefore in the context of non-parametric spatial statistics and its applications. This work makes three major contributions. which are based on the study of non-parametric estimators in a recursive spatial/space-time and revolves around the recursive kernel density estimate in a spatial context, the recursive kernel density estimate in a space-time and recursive kernel regression estimate in space.

Perhaps the most flexible synopsis of a database is a uniform random sample of the data; such samples are widely used to speed up the processing of analytic queries and data-mining tasks, to enhance query optimization, and to facilitate information integration. Most of the existing work on database sampling focuses on how to create or exploit a random sample of a static database, that is, a database that does not change over time. The assumption of a static database, however, severely limits the applicability of these techniques in practice, where data is often not static but continuously evolving. In order to maintain the statistical validity of the sample, any changes to the database have to be appropriately reflected in the sample. In this thesis, we study efficient methods for incrementally maintaining a uniform random sample of the items in a dataset in the presence of an arbitrary sequence of insertions, updates, and deletions. We consider instances of the maintenance problem that arise when sampling from an evolving set, from an evolving multiset, from the distinct items in an evolving multiset, or from a sliding window over a data stream. Our algorithms completely avoid any accesses to the base data and can be several orders of magnitude faster than algorithms that do rely on such expensive accesses. The improved efficiency of our algorithms comes at virtually no cost: the resulting samples are provably uniform and only a small amount of auxiliary information is associated with the sample. We show that the auxiliary information not only facilitates efficient maintenance, but it can also be exploited to derive unbiased, low-variance estimators for counts, sums, averages, and the number of distinct items in the underlying dataset. In addition to sample maintenance, we discuss methods that greatly improve the flexibility of random sampling from a system's point of view. More specifically, we initiate the study of algorithms that resize a random sample upwards or downwards. Our resizing algorithms can be exploited to dynamically control the size of the sample when the dataset grows or shrinks; they facilitate resource management and help to avoid under- or oversized samples. Furthermore, in large-scale databases with data being distributed across several remote locations, it is usually infeasible to reconstruct the entire dataset for the purpose of sampling. To address this problem, we provide efficient algorithms that directly combine the local samples maintained at each location into a sample of the global dataset. We also consider a more general problem, where the global dataset is defined as an arbitrary set or multiset expression involving the local datasets, and provide efficient solutions based on hashing.

Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.

Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.

Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.

The aim of this master's thesis is to design a language through which it would be possible to monitor a stream of network flows in order to detect network policy violations in the local network. An analysis of the languages used in the data stream management systems and an analysis of tasks submitted by the potential administrator were both carried out. The analysis specified resulted in the language design which represents pipelining consisting of filtering and aggregation. These operations can be clearly defined and managed within security rules. The result of this thesis also results in the Policer modul being integrated in the NEMEA system, which is able to apply the main commands of the proposed language. Finally, the module meets the requirements of the specified tasks and may be used for further development in the area of monitoring network policies.

Une grande partie de ces données volumineuses ont plus de valeur lorsqu'elles sont analysées rapidement, au fur et à mesure de leur génération. Dans plusieurs scénarios d'application émergents, tels que les villes intelligentes, la surveillance opérationnelle de grandes infrastructures et l'Internet des Objets (Internet of Things), des flux continus de données doivent être traités dans des délais très brefs. Dans plusieurs domaines, ce traitement est nécessaire pour détecter des modèles, identifier des défaillances et pour guider la prise de décision. Les données sont donc souvent rassemblées et analysées par des environnements logiciels conçus pour le traitement de flux continus de données. Ces environnements logiciels pour le traitement de flux de données déploient les applications sous-la forme d'un graphe orienté ou de dataflow. Un dataflow contient une ou plusieurs sources (i.e. capteurs, passerelles ou actionneurs); opérateurs qui effectuent des transformations sur les données (e.g., filtrage et agrégation); et des sinks (i.e., éviers qui consomment les requêtes ou stockent les données). Nous proposons dans cette thèse un ensemble de stratégies pour placer les opérateurs dans une infrastructure massivement distribuée cloud-edge en tenant compte des caractéristiques des ressources et des exigences des applications. En particulier, nous décomposons tout d'abord le graphe d'application en identifiant quelques comportements tels que des forks et des joints, puis nous le plaçons dynamiquement sur l'infrastructure. Des simulations et un prototype prenant en compte plusieurs paramètres d'application démontrent que notre approche peut réduire la latence de bout en bout de plus de 50% et aussi améliorer d'autres métriques de qualité de service. L'espace de recherche de solutions pour la reconfiguration des opérateurs peut être énorme en fonction du nombre d'opérateurs, de flux, de ressources et de liens réseau. De plus, il est important de minimiser le coût de la migration tout en améliorant la latence. Des travaux antérieurs, Reinforcement Learning (RL) et Monte-Carlo Tree Searh (MCTS) ont été utilisés pour résoudre les problèmes liés aux grands nombres d’actions et d’états de recherche. Nous modélisons le problème de reconfiguration d'applications sous la forme d'un processus de décision de Markov (MDP) et étudions l'utilisation des algorithmes RL et MCTS pour concevoir des plans de reconfiguration améliorant plusieurs métriques de qualité de service. / A large part of this big data is most valuable when analysed quickly, as it is generated. Under several emerging application scenarios, such as in smart cities, operational monitoring of large infrastructure, and Internet of Things (IoT), continuous data streams must be processed under very short delays. In multiple domains, there is a need for processing data streams to detect patterns, identify failures, and gain insights. Data is often gathered and analysed by Data Stream Processing Engines (DSPEs).A DSPE commonly structures an application as a directed graph or dataflow. A dataflow has one or multiple sources (i.e., gateways or actuators); operators that perform transformations on the data (e.g., filtering); and sinks (i.e., queries that consume or store the data). Most complex operator transformations store information about previously received data as new data is streamed in. Also, a dataflow has stateless operators that consider only the current data. Traditionally, Data Stream Processing (DSP) applications were conceived to run in clusters of homogeneous resources or on the cloud. In a cloud deployment, the whole application is placed on a single cloud provider to benefit from virtually unlimited resources. This approach allows for elastic DSP applications with the ability to allocate additional resources or release idle capacity on demand during runtime to match the application requirements.We introduce a set of strategies to place operators onto cloud and edge while considering characteristics of resources and meeting the requirements of applications. In particular, we first decompose the application graph by identifying behaviours such as forks and joins, and then dynamically split the dataflow graph across edge and cloud. Comprehensive simulations and a real testbed considering multiple application settings demonstrate that our approach can improve the end-to-end latency in over 50% and even other QoS metrics. The solution search space for operator reassignment can be enormous depending on the number of operators, streams, resources and network links. Moreover, it is important to minimise the cost of migration while improving latency. Reinforcement Learning (RL) and Monte-Carlo Tree Search (MCTS) have been used to tackle problems with large search spaces and states, performing at human-level or better in games such as Go. We model the application reconfiguration problem as a Markov Decision Process (MDP) and investigate the use of RL and MCTS algorithms to devise reconfiguring plans that improve QoS metrics.

The monitoring of continuous phenomena like temperature, air pollution, precipitation, soil moisture etc. is of growing importance. Decreasing costs for sensors and associated infrastructure increase the availability of observational data. These data can only rarely be used directly for analysis, but need to be interpolated to cover a region in space and/or time without gaps. So the objective of monitoring in a broader sense is to provide data about the observed phenomenon in such an enhanced form. Notwithstanding the improvements in information and communication technology, monitoring always has to function under limited resources, namely: number of sensors, number of observations, computational capacity, time, data bandwidth, and storage space. To best exploit those limited resources, a monitoring system needs to strive for efficiency concerning sampling, hardware, algorithms, parameters, and storage formats. In that regard, this work proposes and evaluates solutions for several problems associated with the monitoring of continuous phenomena. Synthetic random fields can serve as reference models on which monitoring can be simulated and exactly evaluated. For this purpose, a generator is introduced that can create such fields with arbitrary dynamism and resolution. For efficient sampling, an estimator for the minimum density of observations is derived from the extension and dynamism of the observed field. In order to adapt the interpolation to the given observations, a generic algorithm for the fitting of kriging parameters is set out. A sequential model merging algorithm based on the kriging variance is introduced to mitigate big workloads and also to support subsequent and seamless updates of real-time models by new observations. For efficient storage utilization, a compression method is suggested. It is designed for the specific structure of field observations and supports progressive decompression. The unlimited diversity of possible configurations of the features above calls for an integrated approach for systematic variation and evaluation. A generic tool for organizing and manipulating configurational elements in arbitrary complex hierarchical structures is proposed. Beside the root mean square error (RMSE) as crucial quality indicator, also the computational workload is quantified in a manner that allows an analytical estimation of execution time for different parallel environments. In summary, a powerful framework for the monitoring of continuous phenomena is outlined. With its tools for systematic variation and evaluation it supports continuous efficiency improvement.