Fares Grina
Portrait of Fares Grina

I’m an R&D AI engineer at Talan, currently working as a consultant at Coface, where I build AI agents and data pipelines on top of business information and credit-risk data.

Before that, I did a PhD on uncertainty modeling for imbalanced learning with Eric Lefèvre (Université d’Artois) and Zied Elouedi (ISG, Université de Tunis). I used belief function theory to build classifiers that stay reliable when some classes are rare, noisy or overlapping.

These days I think about when we can trust AI systems: how to evaluate LLMs, what their confidence scores actually mean, and when a small decision model is all you need.

EmailLinkedInGitHubScholarX CV (English)CV (français)

News

  • New post: Jev vs. classical ML: how many labels is a good description worth?
  • Our paper Évaluation de la robustesse des LLM : proposition d’un cadre méthodologique et développement d’un benchmark was accepted at the EvalLLM workshop, CORIA-TALN 2025.
  • Presented Evidential Generative Adversarial Networks for handling imbalanced learning at ECSQARU 2023 in Arras, France.

Writing

Research

LLM evaluation, and my PhD work on belief function theory for learning from imbalanced, noisy and overlapping data.

  1. Évaluation de la robustesse des LLM : proposition d'un cadre méthodologique et développement d'un benchmark Fares Grina, N. Kalashnikova EvalLLM workshop, CORIA-TALN 2025
  2. Evidential Generative Adversarial Networks for handling imbalanced learning Fares Grina, Zied Elouedi, Eric Lefèvre ECSQARU 2023 · pdf · code
  3. Re-sampling of multi-class imbalanced data using belief function theory and ensemble learning Fares Grina, Zied Elouedi, Eric Lefèvre International Journal of Approximate Reasoning, 2023 · paper · pdf
  4. Learning from imbalanced data using an evidential undersampling-based ensemble Fares Grina, Zied Elouedi, Eric Lefèvre SUM 2022 · paper · pdf
  5. Evidential hybrid re-sampling for multi-class imbalanced data Fares Grina, Zied Elouedi, Eric Lefèvre IPMU 2022 · paper · pdf
  6. Déséquilibre multi-classes : une approche évidentielle de rééchantillonnage hybride Fares Grina, Zied Elouedi, Eric Lefèvre LFA 2022 · paper · pdf
  7. Uncertainty-aware resampling method for imbalanced classification using evidence theory Fares Grina, Zied Elouedi, Eric Lefèvre ECSQARU 2021 · paper · pdf
  8. Evidential undersampling approach for imbalanced datasets with class-overlapping and noise Fares Grina, Zied Elouedi, Eric Lefèvre MDAI 2021 · paper · pdf