Bonjour à tous,
Pour la rentrée du meetup ML, nous vous proposons une soirée spéciale avec deux talks ! 🚀
Nous serons cette-fois-ci hébergés dans les bureaux de Betclic (117 Quai de Bacalan, 33300 Bordeaux), et vous donnons rendez-vous le mardi 15 Octobre à 19h pour les deux présentations suivantes:
- Learning with Noisy Labels: Challenges and Techniques
Speaker: Mourad INAN (Betclic)
Abstract:
In real-world applications, machine learning models often face the challenge of "noisy labels," which refers to incorrect or inconsistent annotations in the dataset that can degrade model performance. This talk will dive into the difficulties of learning from such imperfect data and discuss various techniques for identifying and mitigating label noise to build more reliable models.
We will explore advanced methods for detecting mislabeled data and improving model training, even when human factors like subjectivity come into play, as is sometimes the case at Betclic, where manual review of alerts by different analysts can lead to inconsistent annotations. As part of this session, we will demonstrate the use of the Cleanlab library, a powerful tool for automatically detecting and correcting label noise in datasets. Attendees will gain insight into practical strategies for managing noisy labels, leveraging Cleanlab to enhance data quality, and improving model robustness in data-centric machine learning environments.
- Advanced RAG: Improve your RAG pipelines by integrating knowledge DBs
Speaker: Vincent Van Steenbergen (ThinkDeepAI)
Abstract:
In this talk, Vincent will present advanced RAG techniques (Multi-agent, GraphRAG, Chain of Thought) that help optimize a RAG pipeline and improve the quality of the provided answers.
Nous nous retrouverons ensuite autour d'un repas / apéritif pour continuer les échanges (fin à 22h). Pour ceux qui le souhaitent, une visite des bureaux de Betclic sera également proposée. 🤗
⚠️ Attention: il est nécessaire de vous inscrire à l'événement pour pouvoir entrer dans le bâtiment. ⚠️
