Yosser Kaddour

Experience

• 

  Centre d'Etude et de Recherche des Télécommunications

  Jun 2022 - Jul 2022

  Summer Internship
• 

  ESPRIT (Ecole Supérieure Privée d'Ingénierie et de Technologies)

  Jun 2023 - Aug 2023

  Data Science Intern

  The primary goal is to leverage the power of deep learning to design a robust and effective model that can remove noise from images, resulting in visually pleasing and high-quality denoised images. The specific objectives include:1 - Designing an architecture: Develop a deep CNN architecture, such as a variant of the U-Net, that is capable of effectively capturing and learning the complex patterns and features present in noisy images.2 - Training data collection: Collect a diverse and representative dataset consisting of clean images and their corresponding noisy versions. 3 - Model training: Train the deep CNN model using the collected dataset to optimize its parameters. Implement appropriate loss functions and regularization techniques to ensure the model learns to denoise images accurately while avoiding overfitting.4 - Evaluation metrics: Establish quantitative evaluation metrics, such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), to measure the performance of the denoising model objectively. Compare the results with state-of-the-art denoising methods to assess the model's effectiveness.5 - Qualitative assessment: Conduct a subjective evaluation of the denoised images to gauge the perceptual quality and visual appeal. Involve human observers who can provide feedback on the denoising results, identifying potential artifacts or improvements needed6 - Enhancements and optimizations: Explore advanced techniques, such as residual connections, skip connections, or attention mechanisms, to enhance the denoising performance. Investigate different network architectures, hyperparameters, and optimization strategies to further improve the model's denoising capabilities.7 - Real-world applicability: Test the trained denoising model on real-world noisy images and assess its performance in practical scenarios. Consider factors such as computational efficiency, real-time processing, and generalization across various types of images. Weniger anzeigen

• 

  Click Erp

  Jul 2023 - Sept 2023

  Machine Learning Intern

  During this internship, I undertook a project focused on developing a real-time vehicle detection, tracking, and counting system using deep learning techniques. The project aimed to address challenges in vehicle detection and recognition, such as variations in vehicle sizes and real-time image collection issues like occlusions and lighting conditions.Key Contributions:Data Collection and Preprocessing: Gathered a diverse dataset of vehicle images from various roads in Tunisia, classified into six distinct classes: car, motorcycle, small truck, big truck, bus, and construction machine. Employed data augmentation techniques to enhance the training dataset.Vehicle Detection: Implemented the YOLOv7 (You Only Look Once) algorithm for efficient and accurate real-time vehicle detection. Fine-tuned the YOLOv7 model to optimize performance for our specific dataset.Vehicle Tracking: Integrated the Simple Online and Realtime Tracking (SORT) algorithm to track vehicles and maintain consistent tracking across video frames.Vehicle Counting: Developed a Python script to automate vehicle counting based on detections and tracks, ensuring each vehicle is counted only once and categorized accurately.Performance Evaluation: Conducted rigorous evaluations and validations to ensure the accuracy of vehicle detection and tracking. Achieved high detection performance and robust tracking results.Project Outcome: Successfully created a system that provides valuable insights into traffic flow, with applications in traffic monitoring and management, enhancing efficiency and safety within the transportation system.This experience allowed me to apply and deepen my understanding of computer vision and deep learning, while also contributing to a practical solution with significant real-world applications. Weniger anzeigen

• 

  Ip-label group

  Feb 2024 - Nov 2024

  Data Science Intern - Performance Graph Interpreter Development

  During my end-of-studies internship at ip-label group , I worked on a project to develop a performance graph interpretation tool for the Ekara platform. This tool aims to enhance client understanding by providing clear and relevant insights into application performance. Key aspects of my role included:Automated Textual Analysis: Developed a system to generate automatic textual descriptions of performance graphs, highlighting trends, anomalies, and key points.AI Predictions: Integrated artificial intelligence to predict potential future issues from graph data, helping clients prevent losses and maximize gains.Custom KPIs: Created innovative KPIs tailored to the specific needs of each client, adding significant value to their performance analysis.Data Visualization and Reporting: Generated intuitive graphs and dashboards to visualize KPIs and important metrics, aiding in decision-making processes.Performance and Scalability: Ensured the tool was responsive and fast, even with large data sets, and designed it to be scalable and adaptable to new metrics and client requirements.Data Security: Implemented measures to ensure the confidentiality and security of client data.This internship provided me with hands-on experience in using artificial intelligence, natural language processing, and data visualization to solve real-world problems in application performance management. Weniger anzeigen

• 

  MSH

  Jan 2025 - now

  Data scientist

  Contribute to the development of SONAR, a behavioral and statistical analysis tool designed to detect fraud patterns in health insurance reimbursements.💡 A few of my key responsibilities: ✅ Designing and improving Machine Learning models to identify unusual and fraudulent behaviors📊 Analyzing data and optimizing detection rules to reduce false positives🤝 Collaborating with fraud analysts and IT teams to ensure model integration and monitoring in production