Pierre-André Noël
Research Scientist
Adaptive Agent
Pierre-André Noël is a Research Scientist at ServiceNow AI Research. His recent work and interests include diffusion models, inference-time efficiency, and/or mechanistic interpretability toward security and reliability.
He holds a PhD in Physics from Université Laval and was a postdoctoral researcher at University of California Davis. His doctoral and postdoctoral research pertained to Complex Networks, Statistical Mechanics and Stochastic Processes. In 2017, he joined an NLP team at Element AI where he spent the following years building core capabilities and working on special projects. ServiceNow acquired Element AI in 2021, and Pierre-André now focuses on fundamental research.
Intérêts
- Relational Models
- Diffusion Models
- Generative Models
- Theory of Machine Learning
Publications
DiffuMamba: High-Throughput Diffusion LMs with Mamba Backbone.
International Conference on Machine Learning (ICML),
2026.
Destruction is a General Strategy to Learn Generation; Diffusion's Strength is to Take it Seriously; Exploration is the Future.
ICLR Blogposts 2026,
2026.
AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding.
Neural Information Processing Systems (NeurIPS),
2025.
Unifying Autoregressive and Diffusion-Based Sequence Generation.
Conference on Language Modeling (COLM),
2025.
Adaptive Diffusion Denoised Smoothing : Certified Robustness via Randomized Smoothing with Differentially Private Guided Denoising Diffusion.
Workshop at the International Conference of Machine Learning (ICML),
2025.
AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Understanding.
Workshop at the International Conference of Learning Representation (ICLR),
2025.
Unifying Autoregressive and Diffusion-Based Sequence Generation.
Workshop at the International Conference of Learning Representation (ICLR),
2025.
BigDocs: An Open and Permissively-Licensed Dataset for Training Multimodal Models on Document and Code Tasks.
International Conference of Learning Representations (ICLR),
2025.
BigDocs: A Permissively-Licensed Dataset for Training Vision-Language Models on Document and Code Tasks.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
Expecting The Unexpected: Towards Broad Out-Of-Distribution Detection.
NeurIPS Datasets and Benchmarks Track (NeurIPS Datasets),
2024.
RepLiQA: A Question-Answering Dataset for Benchmarking LLMs on Unseen Reference Content.
NeurIPS Datasets and Benchmarks Track (NeurIPS Datasets),
2024.
XC-Cache: Cross-Attending to Cached Context for Efficient LLM Inference.
Workshop at the Neural Information Processing Systems (NeurIPS),
2024.
XC-Cache: Cross-Attending to Cached Context for Efficient LLM Inference.
Conference on Empirical Methods in Natural Language Processing (EMNLP),
2024.
Constraining Representations Yields Models That Know What They Don't Know.
International Conference of Learning Representations (ICLR),
2023.
Flaky Performances when Pretraining on Relational Databases.
AAAI-23 Student Abstract and Poster Program,
2023.
Constraining Low-level Representations to Define Effective Confidence Scores.
Workshop at the Neural Information Processing Systems (NeurIPS),
2022.
Exploring the Design Space of Generative Diffusion Processes for Sparse Graphs.
Workshop at the Neural Information Processing Systems (NeurIPS),
2022.
Leveraging Activation Patterns to Define Classifiers Able to Detect and Reject Anomalies.
Montreal AI Symposium (MAIS),
2022.
Flaky Performances when Pre-Training on Relational Databases with a Plan for Future Characterization Efforts.
Workshop at the International Conference on Machine Learning (ICML),
2022.
On the Value of ML Models.
Workshop at the Neural Information Processing Systems (NeurIPS),
2021.