In today's smart world of the Internet of Things (IoT), individuals own their personal set of Internet−connected devices, which collect their private information and personal data for the sake of service provisioning. However, management, aggregation, analysis, and storage of these data are governed by the service providers, beyond the visibility of the owner of the data. By default, the data are retained indefinitely by the service provider. Existing business models do not allow the end−users to have any control over the purpose of use, processing location, and the replication, sharing, and retention policy concerning their personal data. The service providers may use the data for business purposes as per their wish and to their advantages. The goal of my research proposal is poised is to give back the end−users the control over their personal data, and thus, protect the data privacy. The two primary objectives of this research proposal are: (a) giving back the end−users the control over their personal data and (b) implementing the privacy solutions by the principle of design.

Bigdata applications are very resource intensive (mainly memory and IO bandwidth). The naive way to meet their requirements (thus speed up their execution) is to oversize the resources allocated to them. The side effect is that resource over-sizing is the main cause of electricity over-consumption in datacenters. Electricity consumption being the first item of expenditures (represents more that 50%) in the management of a datacenter, operators are faced with two conflicting objectives (performance optimization and energy saving) that should be satisfied at the same time. My research studies address this dilemma. During this seminar, I will present two contributions in this direction.

Urbanization is progressing fast, and it is estimated that by 2050 almost 70% of the total global population will live in cities. This process is expected to bring important advantages, including more efficient running of public services and better living standards for its citizens. However, if not properly managed, it risks aggravating existing issues, such as traffic congestion, environmental pollution, and social inequality. By acquiring, integrating, and analysing large amounts of heterogeneous data, generated in urban spaces by a diversity of sources, such as sensors, devices, vehicles, buildings, and humans, rich knowledge about the functioning of our cities can be derived and used to improve the quality of life of its residents. Drawing inspirations from different fields, including urban planning and economics, Licia will illustrate the models she has built to understand the nature of urban phenomena, with specific applications to inequalities measurement, urban planning, and sharing economy regulation.

L’objectif d’un modèle d’attribution d’auteur consiste à identifier, de la manière la plus fiable possible, le véritable auteur d’un document, extrait d’une œuvre, d’un courriel menaçant ou d’un testament. Récemment, la tétralogie débutant avec L’amica geniale (Une Amie Prodigieuse) a été publié sous le nom de plume d’Elena Ferrante, d’abord en italien puis traduite dans plusieurs langues. Plusieurs noms ont été proposés comme le possible véritable écrivain (par exemple, Milone, Parrella, Prisco, etc.). En s’appuyant sur un corpus composé de 150 romans contemporains italiens écrit par 40 auteurs, trois méthodes d’attribution d’auteur ont été utilisés pour déterminer qui se cache derrière le pseudonyme Elena Ferrante. Dans ce but, la technique du plus proche voisin a été appliquée sur la base des 100 à 2 000 vocables les plus fréquents avec le modèle Delta. Sur la base de cette méthode, on aboutit au nom de Domenico Starnone comme l’auteur derrière Elena Ferrante. Comme deuxième approche basée sur l’ensemble du vocabulaire, le modèle de Labbé confirme cette conclusion. Enfin, la méthode du nearest shrunken centroids (NSC) confirme ce résultat. Pour confirmer ces résultats, une analyse lexicale démontre la proximité entre le vocabulaire de Elena Ferrante et de Domenico Starnone.

Presentation of the recent works in the field of IoT and Blockchain followed by a discussion.

Hyperledger Fabric is an open-source system that provides a modular and extensible platform for deploying and operating permissioned blockchains (distributed ledgers). Starting from the premise that there is no ‘one-size-fits-all‘ distributed consensus protocol, Hyperledger Fabric is the first blockchain system to support pluggable consensus protocols which allows the system to be tailored to particular use cases and trust models. Hyperledger Fabric also features a radically different architecture compared to its predecessors, with the goals of facilitating blockchain performance, scalability, confidentiality and modularity. Unlike other blockchain systems that require their distributed applications to be written in domain-specific languages, Hyperledger Fabric allows the development of distributed applications/smart-contracts in general-purpose programming languages, without dependency on a specific cryptocurrency. Satisfying these requirements required an overhaul of state-of-the-art permissioned blockchain design and rethinking the way blockchains cope with non-determinism, resource exhaustion and performance attacks.

In this talk we discuss Hyperledger Fabric architecture, detailing the rationale behind various design decisions. We also briefly discuss distributed ledger technology (DLT) use cases to which Hyperledger Fabric is relevant, including financial industry, manufacturing industry, supply chain management, government use cases and many more.

The widespread use of mobile phones makes possible to collect mobility traces from users. In the context of project Mobicampus, we aim to leverage this posibility to extract the greatest amount of information. In this work we explore some machine learning methods to infer the transportation mode from mobility traces.

This lecture will address the issue of defining flexible approaches to Information Retrieval by means of the usage of aggregation operators in various phases of the retrieval process. In particular, these approaches rely on the interpretation of Information Retrieval as a Multi-Criteria Decision Making (MCDM) problem, from various perspectives.

Despite the potential impact of aggregation on the whole IR process, this aspect has not received the proper attention in the literature. Only recently some approaches have appeared demonstrating the importance of this issue, and its potential impact on the searching process. This lecture aims to shortly review the main contributions that in the literature have made use of aggregation operators in Information Retrieval.

As opposed to static biometrics, which is related to physical characteristics of human beings, behavioral biometrics identifies people using observations from their activities. As more and more people are using mobile devices for everyday activities, it is obvious to perform the monitoring using these tools.

Password typing rhythm is not a new behavioral biometrics, however the sensors of mobile devices offer new useful features for the study of keystroke dynamics. Not only how we touch the screen is very user-specific, but the micro movements of the device while typing captured by the accelerometer also proved to be very useful.

Several mobile applications are controlled by touchscreen gestures. The device owner can be recognized using these gestures, however, one gesture is not enough for an accurate verification. A few consecutive gestures can predict accurately the user’s identity.

Signatures can now be captured using the fingertip on mobile devices. Not only the trajectory of the signature is captured, but the dynamics of the signing process, which is more difficult to forge.

We review the standard solutions to the above behavioural biometrics. Besides the standard solutions, the proper evaluation methods of such biometrical systems are also presented.

The amount of social events has increased significantly and location-based services have become an integral part of our life. This makes the recommendation of activity sequences an important emerging application. Recently, the notion of a distributed event (e.g. festival or cruise) that gathers multiple competitive activities has appeared in the literature. An attendee of such events is overwhelmed with numerous possible activities and faces the problem of activity selection with the goal to maximise satisfaction of experience. This selection is subject to various uncertainties. In this paper, we formulate the problem of recommendation of activity sequences as a combination of personalised event recommendation and scheduling problem. We present a new integrated framework to solve it and two computation strategies to analyse the categorical, temporal and textual users' interests. We mine the users' historical traces to extract their behavioural patterns and use them in the construction of the itinerary. The evaluation of our approach on a dataset built over a cruise program shows an average improvement of 9.7% over the state-of-the-art.

In the Social Web scenario, where large amounts of User-Generated Content (UGC) diffuse through Social Media often without any form of trusted external control, the risk of running into misinformation is not negligible. For this reason, assessing the credibility of information constitutes nowadays a fundamental issue for users. Credibility, also referred as believability, is a quality perceived by individuals, who are not always able to discern with their own cognitive capacities genuine information from fake one. For this reason, in the last years several approaches have been proposed to automatically assess information credibility in Social Media. Most of them are based on data-driven models, i.e., they employ machine learning techniques to identify misinformation, but recently also model-driven approaches based on Multi-Criteria Decision Making (MCDM) are emerging. Both kinds of models focus on different kinds of features that can be connected to both users and contents. The purpose of this seminar is to illustrate the main concepts, features, and techniques related to the credibility assessment issue, with particular reference to model-driven approaches employing Aggregation Operators to compute, for each information, an overall credibility estimate.

Skyline operator - on Wikipedia

"The Skyline operator is used in a query and performs a filtering of results from a database so that it keeps only those objects that are not worse than any other. This operator is an extension to SQL proposed by Börzsönyi et al. A classic example of application of the Skyline operator involves selecting a hotel for a holiday. The user wants the hotel to be both cheap and close to the beach. However, hotels that are close to the beach may also be expensive. In this case, the Skyline operator would only present those hotels that are not worse than any other hotel in both price and distance to the beach."

Tutorial : exploration de l'algorithme Espérance-Maximisation (EM)

The idea is to define several possible formats of our Friday meetings in order to make them more efficient, not too much time-consuming and at the same time as interesting as possible.

Guido Lena Cota présentera ses travaux de recherche.

Fraud Detection with Machine-Learning and Contextual Knowledge

Fraud Detection with Convolutional Neural Networks (CNN)