Associate Professor and Co-founder of Equall.ai
My Research…
I hold the position of Associate Professor at CentraleSupelec and serve as the Chief Science Officer at Equall.ai, our esteemed legalTech startup. My expertise lies in the forefront of AI advancement, with a particular focus on Natural Language Processing (NLP).
At Equall.ai, we pioneer the development of AI-driven products meticulously crafted for the legal sector, alongside the formulation of innovative workflows designed to streamline lawyers’ tasks, enabling them to work with unprecedented speed and efficiency.
- We developed the first LLM for Law (Colombo et al.)!
At MICS CentraleSupelec - Université ParisSaclay, my research focuses on making AI usable in NLP industrial systems. We publish in top-tiers NLP (ACL, NAACL, EMNLP, AACL,TACL) and general AI (NeurIPs, ICML, TMLR, ACL) conferences and journals. I currently focus on:
- Training large language models and making them efficient for real-world applications.
- The first LLM in French trained from scratch: CroissantLLM (Faysse et al.).
- Achieving near GPT-4 performance on translation and related tasks with a 7B LLM: TowerLLM (Alves et al.).
- A 175B model trained from scratch: BLOOM (Le Scao et al.).
- Making LLM efficient via LLM compression (Boizard et al.).
- Ensuring their safety and robustness.
- Making retriever to abstain via safe text retrieval (Gisserot-Boukhlef et al.).
- Check our works on LLM and generative models for translation (Guerreiro et al., Himni et al), OOD (Darrin et al.) and hallucinations (Guerreiro et al).
- Check our work on safe text classification (Darrin et al.), Out Of Distribution (Colombo et al.) and adversarial attacks (Colombo et al.).
- and developing metrics to ensure accurate measurement of LLM performance.
- Check our work on metrics for text generation (Faysse et al, Guereiro et al., Cchung et al., Staerman et al., Colombo et al.) and our best paper from AAAI (Colombo et al.).
- Check our work on metric aggregation to make sure you can precisely say which LLM is the best (Himmi et al., Colombo et al., Colombo et al.).
I am always happy to discuss new projects and collaborations.