Lotfi Abdi Email & Phone Number

Saint-Denis, Idf, Fr

Saint-Denis, Île-De-France, France

Engineering Manager / Deep Learning \& Machine Learning Perception Systems: Real-time vision-based system for environment perception of Autonomous Train.Led a multidisciplinary team of engineers and data scientists in the development of perception, planning, control, and coordination systems for autonomous train systems.Collaborated with stakeholders to define project requirements and deliver cutting-edge solutions for enhancing the safety, efficiency, and autonomy of train operations.Conducted research and implemented state-of-the-art Deep Learning and Machine Learning algorithms for perception tasks in autonomous train systems.Provided technical guidance and mentorship to team members, fostering their professional growth and ensuring high-quality deliverables.Led the development of a perception system for autonomous train systems, enabling real-time object detection and tracking for enhanced safety and obstacle avoidance.Conducted experiments and optimization efforts to improve the accuracy and efficiency of perception systems for autonomous trains.Collaborated with the engineering team to integrate perception algorithms into the overall control system of autonomous trains.Provide technical leadership, manage the development process, ensure the integration of perception algorithms with other subsystems, and oversee the overall performance, safety, and reliability of the vision-based perception system.

Ville De Paris, Île-De-France, France

Researching and experimenting state of the art machine learning and deep neural networks techniques to predict company financial performance. The objective is to determine a small firm typology and to assess the situation of any company compared to a control group made up of similar companies.Responsibilities: Participated in all phases of data mining; data collection, data cleaning, developing models, validation, visualization and performed Gap analysis.Applied unsupervised and supervised learning methods in analyzing high-dimensional data. Developed and implemented predictive models using machine learning algorithms such as linear regression, classification, multivariate regression, Naive Bayes, Random Forests, K-means clustering, KNN.Evaluated Classification models by Accuracy, Confusion Matrix, Precision, Recall, True Negative Rate (TNR), False Discovery Rate (FDR), ROC — AUC chart.Results:Extensively used Pandas, NumPy, Seaborn, Matplotlib, Scikit-learn, SciPy and Jupyter notebook in Python for developing various machine learning algorithms.Used Pandas, NumPy, seaborn, SciPy, Matplotlib, Scikit-learn, NLTK in Python for developing various machine learning algorithms and utilized machine learning algorithms such as linear regression, multivariate regression, naive Bayes, Random Forests, K-means, & KNN for data analysis.An excellent understanding of both traditional statistical modeling and Machine Learning techniques and algorithms like Regression, clustering, ensembling (random forest, gradient boosting), deep learning (neural networks), etc.Strong problem solving skills; experience dealing with real world large data sets; experience with statistical analysis and machine learning/deep learning.

Région De Grenoble, France

Perception and prediction of pedestrian’ behaviors around autonomous cars using Deep Learning techniques: Human Inspired Autonomous Navigation In Crowds (HIANIC )Description: The HIANIC project try to address the problem of navigating autonomously in shared-space environments, where pedestrians and cars share the same environment.In this project, I focus on the perception and prediction of pedestrians’ behaviors using 3D sensors that are available on the autonomous cars (Depth sensors or Velodyne, with a possible fusion with RGB data).In order to make full use of multimodal information advantages and improve the accuracy and robustness of 3D object detection, I have proposed a LiDAR-camera fusion framework for 3D objectdetection that takes both LiDAR point cloud and RGB images as input and predicts oriented 3D bounding boxes. The primary motivation is to combine complementary information (color,texture and 3D geometry information) sources to enhance the learned representation of the objects and increase the model’s robustness against adverse conditions. First, a stage-1 detection sub-network is trained to generate 3D bounding boxes via segmenting the point cloud into foreground andbackground points. Then, an image segmentation sub-network extracts image features from RGB-image. After that, the 3D proposal generation from points cloud combined with the image features retrieved from the sub-network segmentation output and feed into the stage-2 detection sub-network. Finally, the stage-2 detection sub-network takes the point-wise features augmented from image semantics as input to obtain the final prediction of the 3D bounding box. By fusing informationobtained from different sensors, we observe that the accuracy and recall achieved are higher even for small and distant objects.

Clermont-Ferrand, Auvergne, France

Deep Learning for autonomous vehicles: Reliable and Safe long-term auTonomy for Intelligent Navigation based on Generic approach (tRuSTING)Description The goal of the project is to design and develop a perception-based self-drivingsystem for urban scenarios.I have proposed a novel transfer-learning framework (Spatio-Temporal Specialization for Semantic Video Segmentation) using the spatio-temporal coherence in video data for semantic video segmentation. The objective is to couple the decisions taken by a CNN with the temporal coherence in video images via optical flow to enhance the accuracy of semantic video segmentation.

Région De Clermont-Ferrand, France

Segmentation of brain MRI structures with deep learning: DEEP BRAIN LEARNING (DEEP-BLearning)Description The aim of this project is to implement and make a preliminary evaluation of a method based on deep learning technique to improving automatic deep brain segmentation by using 3D and fully convolutional neural network (3D-FCNN) on a unique dataset composed of expert labelling in different imaging modalities.Inspired by the recent success of dense networks, I have developed an approach to 3D deep brain segmentation based on a volumetric FCNN.

Gouvernorat De Sousse, Tunisia

Lotfi Abdi received the graduate degree in computer science from the Higher Institute of Applied Sciences and Technology of Sousse, University of Sousse, Tunisia, in 2009, the engineer degree in Computer Science and the master degree in Intelligent and Communicating Systems from the National Engineering School of Sousse, University of Sousse, Tunisia, in 2012 and 2013, and the Ph.D. degree in Communication Systems at the National Engineering School of Tunis, University of Tunis El Manar, Tunisia, in 2016. His research interests include Intelligent Transport Systems, Computer Vision, Signal, Image and Video Processing, Augmented Reality, Artificial Intelligence and Neural Networks.

Doctor Of Computer Science, Deep Learning For Autonomous Vehicles

Driver information system: a combination of augmented reality, deep learning and vehicular Ad-hoc networks This thesis dealt.

Master Of Computer Applications (Mca), Ingénierie Informatique

Engineer'S Degree, Computer Science

Graduate Degree, Computer Science