Five master interns will present their research activity along with a Ph.D. student on April 12, 2022, 2-4 pm, in room 37.2.43, 2 rue conté, Paris.

More details are below.
For remote attendance, please write to nour-el-houda.yellas AT cnam DOT fr

Rezak AZIZ

Title: Clustering with missing data in the context of homomorphic encryption

In our days, a huge amount of data are generated every day. This amount is increasing exponentially with the advent of Internet of things. The data collected by IoT devices need to be processed to extract knowledge andexploit it. To do that, generally,  we use Machine learning techniques such as clustering. However,  the IoT devices don’t have  enough resource capacity to do such processing. To avoid processing data within the IoT devices, the trend is to outsource the senseddata to the Cloud that has both resourceful data storage and data processing. Nevertheless, the externalized data may be sensitive, and the users may lose privacy on the data content while allowing the cloud providers to access and possibly use these data totheir own business. To avoid this situation and preserve data privacy in the Cloud datacenters, one possible solution is to use the fully homomorphic encryption (FHE) that assures both confidentiality and efficiency of the processing. Indeed, the challengingissue in this context is to adapt the clustering approaches to apply them on encrypted data so that the results of calculus over encrypted data can be reported on the clear-text data. Several works have studied clustering algorithms in the context of homomorphicencryption, but, all these works did not take into account the problem of missing data. This problem is ubiquitous in reality and has been studied in clear text. In our work, we aim to propose a clustering solution using k-means in the context of homomorphic encryption while handling missing data.

Rezak AZIZ is an intern at CNAM under the supervision of Samia Bouzefrane and Vincent Audigier. He is a final year student at Higher national school of computer science (ESI ex INI) in Algiers, where he is preparing a master and engineering degrees in computer systems field.

Leticia SADI

Title : Trust Evaluation in IoT environments using Interpretable Models 

The Internet of Things (IoT) is a field that has experienced explosive growth in recent years. It allows the collection of massive amounts of data through millions of interacting devices. Decision making is therefore delegated to the connectedobject systems, which minimizes human intervention and reduces costs. Despite all these advantages, IoT objects have many constraints that make them vulnerable to many risks. As the application of traditional security methods is difficult in the IoT context,trust evaluation is becoming more and more important.
Trust assessment is the process of quantifying the trust of network nodes. Trust revolves around the assurance that people, data, entities, or processes will function or behave in the expected manner. In an IoT environment, trust assessmentconsists in establishing a belief in the reliability of a network object. This trust is performed according to the characteristics and the behavior of the IoT objects.
In this talk, I will first introduce trust concept in IoT environments, then I will discuss trust assessment using machine learning. I will also discuss the importance of interpretability of machine learning in such a sensitive process.  

Leticia SADI is a fifth year student at the Ecole Nationale Supérieure d’Informatique (ESI) of Algiers, specializing in computer systems.  She is preparing a double diploma: an engineer degree and a master’s degree in computerscience for her graduation this year 2021/2022. For her Engineer/Master internship, she joined the ROC team to work under the co-supervision of Samia Bouzefrane & Ndeye Niang. 


Title: GAN based RSS data recovery for indoor localization

With the growing emergence of the Internet of Things and the importance of position information in this context, localization is attracting more and more attention in the researchers’ community. The outdoor location is provided by GPS which is not suitable for indoors environments. In general, conventional Indoor Positioning Systems (IPSs) either adopt geometric mapping approach or resort to the Location Fingerprinting (LF). For geometric mapping, intermediate spatial parameters like distance or direction are first derived from certain physical measurements. Then, target’s physical location can be further inferred by using geometric algorithms (e.g., trilateration or triangulation). As an emerging indoor positioning, location fingerprinting benefits from a pattern-matching mechanism, which comprises offline training phase and online location estimation phase. Specifically, in the offline phase, wireless signatures are collected at a set of geo-tagged Reference Points (RPs) in the area of interest to construct the fingerprint database. During the online phase, the measured signature at an unknown position is matched with the offline radio map to return the best-fitted location estimation. In both approaches Received Signal Strength (RSS) as the indicator of Medium Access Control (MAC) layer’s link quality, is used. The rich indoor multipath and shadowing effects led on missing RSS measurement during localization process and thus degenerate positioning accuracy.
Advanced optimization algorithms have been used for completing the inter-node distance matrix from partially known data. However these methods have shown their limits. We propose in this work to implement Generative Adversarial Network GAN to perform Matrix completion to enhance localization performances.

Serbouh Céline is an intern at Cedric laboratory of CNAM, her internship aims to implement the GAN method to improve the performance of indoor localization in order to get her data-processing engineer degree from the ESI where she recently got a master degree on.
Passionate about machine learning, her main professional objective is working on a PhD thesis in her favorite field.


Title: Data Management for Intelligent Decision-Making in Digitalized Processes

Industry 4.0 marks a new industrial revolution allowing more connection of the virtual and the real world. With implementation of Industry 4.0 concept, most of the working environment gets automatized with Cyber-physical systems and Internet of Things with data being processed online using cloud computing. These technologies stimulate incremental changes in processes over time. This phenomenon implies the process variability, i.e., different instances of the same process during its execution can take particular ways depending on the context. We foresee the problem of processes variability as a decision-making (DM) problem as each process contains variation points in which the next process step should be selected between at least two alternative paths. With faster and more complicated processes, bigger amounts of data need to be handled to make decisions. The challenge is to manage these huge amounts of data in a digitalized environment.
The main objective of this work is to propose an ML-based approach for intelligent decision-making to deal with process configuration, optimization, and prognosis in the current context of Industry 4.0.

Lilia MEZANI is an intern at CNAM under the supervision of Elena Kornyshova. She is preparing her engineering degree at the Higher National School of Computer Science (ESI ex INI) in Algiers, Algeria. The main objective of her internship is to propose an ML-based approach for intelligent decision-making to deal with process configuration, optimization, and prognosis in the current context of Industry 4.0.


Title : Differential privacy in the context of federated learning and collaborative  clustering 

The volume and variety of data is growing exponentially due to various applications of computers in all fields. This growth has been achieved through the affordable availability of computer technology, storage and network connectivity. The collection of digital information by governments, businesses, and individuals has created enormous opportunities for knowledge-based decision making and information. However, in their original forms, data generally contain sensitive information about individuals, and their publication is a violation of their privacy. 
It makes sense then that the advent of Big Data, machine learning and data science requires a reconsideration of privacy. In fact, privacy-preserving machine learning has been extensively studied and many anonymization techniques have been proposed. However, the dependence of currently available privacy models on the adversary’s contextual knowledge makes privacy protection difficult. Hence the emergence of the differential privacy approach.
Differential Privacy (DP) is a branch of statistics that attempts to ensure the protection of sensitive information about an individual, regardless of the attacker’s background knowledge. It’s a rapidly growing concept that is currently used in a wide range of aspects, including federated learning. Despite the fact that  federated learning offers a privacy-aware paradigm that does not require data sharing; it does not always provide sufficient privacy guarantees due to the uploading of model parameters. Hence the need to use a differential privacy in this context. 
While federated learning has been studied with DP in many research works, there is no work that investigates collaborative clustering with differential privacy.
The aim of our work is to give an extensive investigation of differential privacy when combined with federated learning and the way DP can be employed to secure collaborative clustering. 

Saloua BOUABBA  is currently a student in the last year at Ecole Nationale Supérieure d’Informatique (ESI) in Algiers, with the objective of obtaining in september  2022 a double degree of Engineer and Master in computer science. She joined the CNAM of Paris to do her final year research internship under the co-supervision of Samia Bouzefrane and Ndeye Niang. 


Title: Intent-based configuration of information and communication components for Industry 4.0 applications.

Industry 4.0 is rife with challenges and opportunities. There is no end to what this digital revolution has brought to factory settings and beyond, starting with the integration of cyber-physical systems, augmented reality, the Industrial Internet of Things (IIoT), cloud computing and so on. Thus, companies are still struggling to evolve their business intentions and the ICT configuration.
Intent-based approaches apply a deeper level of intelligence to perform routine tasks, set policies, respond to system events, and verify that goals and actions have been achieved. For example, Intent-Based Networking(IBN) a paradigm shift in traditional networking allowing network managers/users to express the configuration required, we can mention Cisco SD-Access as the first in its domain to present an IBN solution for the enterprise. Our goal is to capitalize on the steps of this innovation to respond to I4.0 challenges with a conceptual framework based on intents.
In this presentation, we will briefly present the context of our research project and the problem we are addressing. Then we will explain the concept of intent-based approach relying on the architecture of IBN; finally, we will talk about the solution from a high-level perspective.

Kaoutar SADOUKI earned her Master’s double degree in 2021 in computer science from Sidi Mohamed Ben Abdellah University and Sorbonne University. Her current research focuses on leveraging Intent-based techniques to design and build solutions that address both new difficulties and corporate objectives while applying ICT components of industry 4.0 applications.

ROC interns seminar – April 12, 2022
Tagged on:     

Recent publications