Ruben Bousbib

• Entraînement distribué de modèles de reconnaissance de la parole state-of-the-art. • Déploiement des modèles. • Optimisation des modèles.

NLP-inspired software fuzzing - Supervised by Lucas Hosseini and Edouard Grave. • Developed a fuzzing dataset generation tool to collect interesting mutations of compiled programs. • Implemented from scratch a distributed training pipeline and Transformer-based models to learn byte-level code coverage information.

Spoken Language Understanding – Supervised by Prof. Yannick Estève and Alice Coucke. • Developed multimodal models and ML experiments for Addressee Detection on privacy-aware and embedded voice assistance technologies. • Implemented the data generation pipeline and deployed to production.

Cognitive Machine Learning (CoML) team - Supervised by Prof. Emmanuel Dupoux (FAIR). • Implemented an approach for domain adaptation of end-to-end voice activity detection. • Developed deep multi-task models for voice type classification from the raw waveform. • Designed and run the experiments on a computer cluster and published results in top conference.

Software Engineerging and Data Analysis to redesign the payroll system of the French armed forces. • Software development in Java. • Bash scripting to monitor data warehousing.