🎓 Education
- 2023–Present — Machine Learning Master of Research, MILA/ETS, Montréal
- 2021–2022 — Master Maths Vision Apprentissage, ENS Paris-Saclay
- 2019–2021 — Master of Fundamental Mathematics / Agrégation de Mathématiques, ENS Rennes
📄 Research
- Boammani Aser Lompo Making Learned Programs Explicit: direction-level exclusivity as a safety criterion for editing LLM circuits. Ongoing — Project page · GitHub
- Boammani Aser Lompo, Marc Haraoui
MODALITY-SWAP DISTILLATION: Rendering Textual Reasoning into Visual Supervision.
(submission soon) — PDF - Boammani Aser Lompo, Marc Haraoui
Visual-TableQA: Open-Domain Benchmark for Reasoning over Table Images.
NeurIPS 2025 Workshop on Foundations of Reasoning in LMs — arXiv - Boammani Aser Lompo, Patrik Kenfack Implicit Bias of Gradient Descent under Feature-Mediated Spurious Correlations. Preprint, under submission — Page · GitHub
- Camille-Sovanneary Gauthier, Romaric Gaudel, Elisa Fromont, Boammani Aser Lompo Parametric Graph for Unimodal Ranking Bandit. ICML 2021 — HAL
- Boammani Aser Lompo, Thanh-Dung Le, Philippe Jouvet, Rita Noumeir
Multi-objective Representation for Numbers in Clinical Narratives: A CamemBERT-Bio-Based Alternative to Large-Scale LLMs. under submission — doi - Reviewer: ICCV 2023 • CVPR 2026 • NeurIPS 2026 • ICLR 2026
🧩 Featured Projects
🔍 Making Learned Programs Explicit
Mechanistic interpretability · GPT-J / Llama-3-8B · Circuit ID + SAE feature tracking · Edit-safety criterion
🛡️ Youth Mental Health Safety Guardrail
Fine-tuned Qwen2.5-7B + LoRA · 9-signal distress taxonomy · F1=0.85
🧾 Visual-TableQA Pipeline
Multimodal reasoning · 2 500 LaTeX tables · 9 000 QA pairs · LLM-jury QC
🧬 Multi-Objective Token Representation
LESA + Xval · Numerical reasoning · CamemBERT-Bio
⚖️ Optimal Transport for Color Transfer
Unbalanced OT · Color distribution transfer · Visualizations
♟️ Reinforcement Learning for Chess
C++ · SARSA + Monte Carlo · ~1350 Elo after 1h self-play
🎯 Interests
- AI Safety & Mechanistic Interpretability — circuits, sparse autoencoders, and safe model editing
- Learning theory (implicit bias, spurious correlations, generalization)
- Applied Mathematics, AI/ML (Reasoning, Foundation Models, Reinforcement Learning), AI Fairness
💼 Internships
- CREST (ENSAI, Rennes) — 2021
Unimodal Bandits for Ranking Systems (Recommenders, MAB, Learning to Rank) . Supervisor: Romaric Gaudel - COMETE Lab (Inria/École Polytechnique, Paris) — 2020
Measuring privacy of cryptographic systems; optimal privacy properties. Supervisor: Catuscia Palamidessi
🌟 Distinctions
- Best Oral Presentation, 91e Congrès de l’Acfas (2024)
- Scholarships: R-SADC research grant (23k), French Government Excellence Grant (50k)
- Math: IMO Honourable Mention (2014) • PAMO Gold (2015) • National winner (Burkina Faso, 2015)