Salim Aissi
🎓 PhD student in AI · ISIR, MLIA team · Sorbonne Université, Paris
ISIR, Sorbonne Université
Paris, France
👋 I’m a PhD student in artificial intelligence at Sorbonne Université, working at ISIR (Institut des Systèmes Intelligents et de Robotique) in the MLIA team, under the supervision of Prof. Laure Soulier, Prof. Olivier Sigaud and Prof. Nicolas Thome.
🤖 My research is on multimodal agents for robotics and embodied AI: how vision-language models and large language models can perceive, reason and act in the physical world. I work on generative models and sequential decision making.
📄 Recent work covers aligning LLM agents with reinforcement learning (NAACL 2025 Findings), coupling perception and reasoning for embodied decision making (PRISM, ICML 2026), and restoring the language abilities of vision-language-action models through parameter merging (EACL 2027).
🏫 Before my PhD, I graduated from the Master IMA (Image) at Sorbonne Université and did a research internship in computer vision with Prof. Isabelle Bloch at LIP6.
news
| Nov 15, 2026 | 🎉 Bring Back Language to Actions: Restoring VLM Capabilities in VLA Models via Parameter Merging was accepted at EACL 2027 (Main Conference)! arXiv version coming soon. |
|---|---|
| Sep 30, 2026 | 🗼 Our paper Bring Back Language to Actions was accepted at the NeurIPS 2026 Workshop RoboPAD (Post-Training Adaptation of Robot Foundation Models), at the NeurIPS Paris venue! |
| Jul 01, 2026 | 🇰🇷 Attended ICML 2026 in Seoul, South Korea, to present PRISM. |
| May 06, 2026 | 🎉 PRISM: Perception Reasoning Interleaved for Sequential Decision Making was accepted at ICML 2026! arXiv |
| May 01, 2025 | ✈️ Attended NAACL 2025 in Albuquerque, New Mexico, to present our Findings paper on prompt overfitting in RL-aligned LLM agents. |
| Mar 01, 2025 | 🏔️ Attended the ALPS Winter School (Advanced Language Processing School). |
| Jan 22, 2025 | 🎉 Our paper Reinforcement Learning for Aligning Large Language Models Agents with Interactive Environments: Quantifying and Mitigating Prompt Overfitting was accepted at NAACL 2025 (Findings)! arXiv |
| Jan 08, 2024 | 🎓 Started my PhD in artificial intelligence at ISIR, Sorbonne Université (MLIA team). |
selected publications
- Bring Back Language to Actions: Restoring VLM Capabilities in VLA Models via Parameter MergingIn Proceedings of the Conference of the European Chapter of the Association for Computational Linguistics (EACL), Main Conference, 2027Also accepted at the NeurIPS 2026 Workshop RoboPAD (Paris). arXiv coming soon