About

I work on mechanistic interpretability, focusing on how language models handle knowledge and process information. My research particularly investigates tricky situations such as conflicts between learned parametric knowledge and contextual knowledge, with the goal of mitigating hallucinations in large language models.

I enjoy building practical ML systems end‑to‑end: data, modeling, evaluation, and deployment, with a particular focus on understanding the internal mechanisms of neural networks.

Education

PhD, Natural Language Processing
BNP Paribas & Sorbonne University

Experience

PhD Researcher
BNP Paribas · Sorbonne University
Mechanistic interpretability; knowledge handling in language models; hallucination mitigation

Research Focus

Mechanistic Interpretability
Understanding internal mechanisms of language models
Developing techniques to understand how language models process and integrate different sources of knowledge.
Knowledge Source Attribution
Parametric vs Contextual Knowledge
Investigating conflicts between learned parametric knowledge and contextual information to mitigate hallucinations.
Probing Techniques
Model Analysis and Evaluation
Developing probing methodologies to understand model behavior and knowledge processing mechanisms.

Publications

Z Tighidet, A Mogini, J Mei, B Piwowarski, P Gallinari · BlackBoxNLP@EMNLP 2024
Z Tighidet, L Labiod, M Nadif · SFC2023 · 2023

Contact

The fastest way to reach me is via LinkedIn.