PhD Position in Scientific Claim Verification in Research Articles

This PhD position at Inria’s Rennes center focuses on advancing AI techniques to verify scientific claims by identifying, validating, and explaining evidence across text, tables, figures, and citations. The project aims to address critical challenges in AI-assisted scientific writing, where large language models may introduce unsupported or misleading claims. The successful candidate will design open benchmarks, develop multimodal verification methods, and evaluate robustness and evidence use. The work is embedded in the ANR Fabuleux project, offering interdisciplinary collaboration in a dynamic research environment.

• Develop open benchmarks for multimodal scientific claim verification, including text, tables, figures, and citations.
• Design and implement methods to identify and align scientific claims with supporting evidence across multiple modalities.
• Evaluate verification methods for accuracy, robustness, and evidence use, ensuring decisions are grounded in reliable sources.
• Produce scientific writing (papers, technical reports) to disseminate findings and methodologies.
• Collaborate with interdisciplinary teams to integrate verification tools into AI-assisted scientific writing workflows.
• Conduct experiments to build and annotate datasets for training and validating verification models.
• Analyze model behavior critically to improve reliability and explainability of AI decisions.