This paper presents a study of various standard anomalies detection techniques in internet networks, using machine learning algorithms, classification algorithms, and graph techniques working on SQL/Mapreduce and SQL-Graph implementation platforms. This approach shows its efficiency for the study of large datasets. Firstly, several algorithmic approaches and possible implementations have been studied and experimentally tested to see how the SQL-MapReduce and SQL-Graph implementation can process large-scale data. Secondly the performance and scalability of the algorithms for large volumes of data have been compared to choose the most appropriate for typical anomaly detection. The application scope is very broad, such as spam detection, crime detection, mafia or terrorism community detection, network intrusion detection, malignant tumors detection in healthcare, fraud detection on banking transactions, identity theft detection etc. The main problem we will address on this paper is the processing performance on large scale data using an implementation of Massively Parallel Processing (MPP), SQL-MapReduce, and SQl-Graph.

Social network sites with large-scale public networks like Facebook, Twitter or LinkedIn have become a very important part of our daily life. Users are increasingly connected to these services for publishing and sharing information and contents with others. Social network sites have therefore become a powerful source of contents of interest, part of which may fall into the scope of interests of a given group. So far, no efficient solution has been proposed for a group of interest to tap into social data, especially when they are protected by and scattered across different social network sites. We have therefore proposed a user-centered approach for integrating social data into groups of interests. This approach makes it possible to aggregate social data of the group's members and extract from these data the information relevant to the group's topic of interests. Moreover, it follows a user-centered design allowing each member to personalize his/her sharing settings and interests within their respective groups. We describe in this paper the conceptual and technical components of the proposed approach. To illustrate further the approach, a web-based prototype is also presented. A preliminary test using this prototype was carried out and showed encouraging results.

Online social networks (OSNs) such as Facebook, Twitter, Linked In, Google+, have become extremely popular and ubiquitous today. Users are actively connected to these services for creating and sharing contents and events with others, and, in some cases, this activity takes place in the scope of groups of interests. Therefore, from amongst the morass of data generated every day by users, a part of information may match the interests of certain groups. In practice, members are not all linked to each other via each OSN they are connected to. It is also not realistic to assume that each member can manually explore all others' social profiles to reach the information that may be relevant to their interests. Thus, there is a need for aggregating members' social streams on a single information support to collect relevant information, and, consequently, to promote collaborative knowledge-sharing. However, the disconnected nature of today social websites prevents a straightforward aggregation process. An efficient automated aggregation model is needed. We present, in this paper, the idea of empowering collaborative intelligence by the use of a user-centered approach for OSN aggregation. We illustrate the approach by a first experience to evaluate its impact on users information sharing and enrichment capabilities.

This work presents a method that aims at structuring the design process of interactive information systems (IIS). This method, which takes parts of its ground in the joint cognitive systems approach, formalizes the integration of usage data in the design process of IIS. Data are generated by using methods of activity analysis. The assumption is that for integrating usage data successfully, they must be represented in a formalism close to the development language, while maintaining the formalization language accessible to each domain specialist involved in the design process. The method can be split up into in five steps: (a) data collection of users' needs in different contexts, (b) representation of the collected data, (c) modeling of corresponding knowledge in an implementation formalism, (d) specification of the IIS, and (e) validation of the model by a functional specification aid tool. A functional specification design tool has been developed according to the presented approach, and has been applied to the design of a traveler information system.

Hierarchical penalization is a generic framework for incorporating prior informa- tion in the fitting of statistical models, when the explicative variables are organized in a hierarchical structure. The penalizer is a convex functional that performs soft selection at the group level, and shrinks variables within each group. This favors solutions with few leading terms in the final combination. The framework, orig- inally derived for taking prior knowledge into account, is shown to be useful in linear regression, when several parameters are used to model the influence of one feature, or in kernel regression, for learning multiple kernels. Keywords – Optimization: constrained and convex optimization. Supervised learning: regression, kernel methods, sparsity and feature selection.

The research activity of the Heudiasyc laboratory in the domains related to WIC concerns four projects : (i) intelligent agents, (ii) e-learning, topic maps, and semantic Web, (iii) Web topology, and (iv) digital music preservation.

This paper presents a framework to build hypertext system-based diagnosis. We first propose a definition of hypertext systems, which is more appropriate to account for the structural properties that exist in any document. These structures become perceptible by the means of projections. A projection is decomposed into two steps: an interpretation operation which is the computation of the document, and an instantiation operation which is the transformation of the digital document into a perceptible document. Next, we show that by separating nodes content from hypertext documents structures, it becomes possible to implement efficient tools for document reading and synthesis. Finally, we refer to a model ontology building method which we use for building domain dependant diagnosis ontology to model hypertext-system based diagnosis. The project of building a hypertext-based system for aiding in the acoustic and vibration diagnosis of rotating machines is used to discuss a possible application of the system.

Research issues in the domain of AI for design can be organized in three categories: decision making, representation and knowledge handling. In the area of knowledge handling, this paper addresses issues concerning the management of design experience to guide a priori the generation of candidate solutions. The approach is based on keeping the trace of a previous design experience as a hierarchical knowledge base. A level in the hierarchy can be viewed as a level of granularity of the description of the design process. A general framework for defining a partial order function between the granularity levels in the knowledge bases of design expertise is proposed. It is then possible to compute the sets of the elements belonging to smaller granularity levels, which are linked to any component of the hierarchy. Thus, it makes it possible to compute the level in the hierarchy that can be reused without modification for the design of a new product. Computation of the appropriate level is mainly based on matching the data corresponding to the new requirements with these sets. The approach has been tested by using a multiple expert systems structure based on using interactively two systems, an expert system development tool for design, KAUS, and an expert system development tool for diagnosing engineering processes, SUPER. The intrinsic properties of SUPER have also been used for improving the design procedure when qualitative and quantitative knowledge is involved.>

A method of managing and updating a database of facts for engineering systems that change their states over time is proposed. Plan selection and monitoring of dynamically evolving processes involves: (1) deciding whether or not action is warranted based on information about the current state of the system, (2) choosing an appropriate set of actions to change its state, and (3) inferring the effect of those actions on the system's state. The proposed methodology is based on the assumption that the monitored system follows a logical model. An intuitive definition of this logical model and a short justification of this assumption are presented. The knowledge-acquisition system (Super) that was developed to build a network of rules that represents this logic model and the properties of the knowledge-base structure are briefly presented. A strategy to maintain and update the database is defined. Data are correctly inserted, deleted, or changed by using the definition of consistency based on the network of rules. The case of rules that represent the changes of states of the system in response to external actions is developed. Two solutions are presented, depending on whether the history of the state changes is maintained or not. A looped inference engine is introduced to ensure goals analysis and plan selection according to the past and present data. The strategy is based on assigning a goal a rating depending on two factors: the ratio of the number of supporting facts to a function of the total number of facts in the database, and the relationships of the supporting facts to the goals.>

An extension of the method that was used for building and using the network-based knowledge system SUPER is proposed to fully utilize the benefits of this approach in the domain of diagnosing distributed dynamically evolving processes. In the scope of distributed sensor networks, several issues are addressed concerning problems dealing with software architectures, strategies, and properties for efficient sensor data management: (1) how to build links efficiently between the elements of each network-based knowledge base, (2) how to maintain consistency of the whole structure and manage the constraints of domain dependent variables corresponding to sensor data, (3) how to manage and update efficiently the database at run time in order to maintain its consistency and to satisfy a high level of response time performances, and (4) how to propagate the solutions given by a qualitative knowledge base into a knowledge base utilizing sensor data whenever the sensors are out of order. The answers given are based on extending and generalizing the principles that have been defined for SUPER.>

The problem of using global variables to improve the representation capacities of the SUPER expert system is addressed. The first difficulty is to propagate the domain of definition of the variables along the knowledge base using the AND/OR transitive relations between rules. The second problem is to upgrade the algorithm that has been defined for the consistency maintenance of the knowledge base for propositional logic to the level of logic with global variables. Answers are given to both problems: (1) the principles for chaining rules with global variables are given; (2) tools for computing the propagation of the domain of variables along the knowledge base are defined; (3) consistency definitions are given and tools for maintaining the consistency of the knowledge base as it incrementally increases are proposed.>