Mes compétences :
Linux
Python
Localization
Software Engineering
5G
C
Tensorflow
Research
Keras
Computer Vision
Deep Learning
Java
Git
SLAM
Machine learning
C++
Entreprises
Télécom Sudparis/ Upmc Paris 6
- PHD Researcher
2017 - maintenantPhD in 5G Localization using Machine Learning and AI. My deep learning based localization solution using CSI of MIMO antennas attained a 2.3 cm accuracy winning the first place in the CTW 2019 indoor positioning competition (https://ctw2019.ieee-ctw.org/authors/#databakeoff).
During PhD, I taught a total of 52 hours for different modules concerning Algorithmics, Unix, NS-3 Network Simulator for Msc and Bachelor students both in English and French. Also, I was elected as PhD representative of the Ecole Doctoral STIC for a year and a half.
My PhD research is concducted under the supervision of Prof Eric Renault of Telecom Sudparis and Prof Paul Mühlethaler of INRIA.
Siemens
- Msc Engineer
Saint-Denis2016 - 2016A 6-months internship to work on Msc project and thesis. The topic of my work is Graph-based Localization with Velodyne VLP-16. Data from Velodyne's VLP-16 laser sensor is processed to localize a robot in an unknown environment. This is achieved through the use of factor graphs in gtsam library (C++). Edge points landmarks are extracted as in the paper: "LOAM: Lidar Odometry and Mapping in Real-time". The robot's pose is estimated probabilistically by re-observing landmarks between two poses and resolving the factor graph. The contribtuions can be summarized into the following points:
1- Organized point cloud of VLP-16 sensor for fast indexing.
2- Introduced a method to remove non-robust edge point features to mitigate the effect of wrongly associated features on pose estimation.
3- Introduced 2 factors for factor graph pose estimation in GTSAM library.
Télécom Sudparis/ Upmc Paris 6
- Research Intern
2015 - 2015In the scope of reputation systems, the effect of forceful peers (members trying to affect other members to follow their opinion) in a network is studied in the context of elections; each forceful peer is considered as a candidate in an elections and connects into the network with a defined budget of edges. Different strategies to connect into a network are studied and compared by applying Degroot-model. Simulations are endowed for different graph topologies (Erdos-Renyi, geometric, Barbasi-albert) with different strategies.
A smart forceful peer is introduced who uses proir knowledge of opponent's connections to win with minimum possible connections. This is achieved by forming an MILP minimization problem and solving it using Gurobi solver in python.
A paper of this work "How to win Elections" was published in Collaboratecom 2016 conference in Bejiing, China.
• PhD: 5G Localization in different contexts.
• Machine Learning/ Deep learning-based localization using CSI of a MIMO antenna; won the first place in the CTW 2019 indoor positioning competition (https://ctw2019.ieee-ctw.org/authors/#databakeoff).
• Range-based localization triangulation.
• Courses and summer schools ~250 hrs in prestigious institutes; Ecole Polytechnique, Siemens, Universite
