Yohan Chatelain

High-performance computing • Scientific software systems • Numerical reliability
Montréal, Quebec, Canada  |  yohan.chatelain@gmail.com  |  +1 514 206 2468
Website    Google Scholar    GitHub    ORCID 0000-0001-7023-250X    LinkedIn

Research Profile

Researcher developing compilation, execution, and statistical analysis methods to make large-scale scientific computing faster, more reproducible, and numerically reliable. My research connects high-performance computing (HPC), software systems, and floating-point computation analysis: LLVM instrumentation, heterogeneous CPU/GPU execution, vectorized stochastic arithmetic, mixed and adaptive precision, parallel scientific workflows, and reliability testing for AI and biomedical pipelines. My research program aims to develop an open, end-to-end software stack capable of observing, quantifying, and controlling numerical uncertainty on heterogeneous HPC systems while improving performance, memory footprint, and energy efficiency.

Academic and Research Experience

Scientific Associate

Toronto, Canada • –present
Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH)
  • Scientific associate for the Brain Health Data Challenge platform, which supports safe, responsible, and reproducible AI research on sensitive health data; developing computing infrastructure and evaluation methods for collaborative predictive modelling challenges.
  • Analyzing the numerical stability of neuroimaging models, including feature extraction with untrained CNNs and longitudinal analysis, to assess their robustness and reproducibility.

Postdoctoral Researcher

Toronto, Canada • –
Krembil Centre for Neuroinformatics, CAMH
  • Quantified the numerical variability of MRI-derived biomarkers and propagated uncertainty through to the scientific conclusions of Parkinson's disease studies.
  • Created PRISM and developed Fuzzy PyTorch, reducing stochastic-arithmetic instrumentation time by up to 60× compared with binary-instrumentation approaches and evaluating models with up to 341 million parameters.

Postdoctoral Researcher

Montréal, Canada • –
Department of Computer Science and Software Engineering, Concordia University
  • Led research at the intersection of HPC, numerical analysis, neuroimaging, bioinformatics, and AI; developed PyTracer, Fuzzy, and statistical stability tests for scientific pipelines.
  • Coordinated collaborations with McGill University, Université de Montréal, the University of Washington, Stanford University, and European partners; mentored students at the undergraduate, master's, and doctoral levels.

Doctoral Researcher

Versailles, France • –
Université Paris-Saclay / UVSQ, LI-PaRAD
  • Developed LLVM-based tools for numerical debugging and reduced-precision optimization in HPC: Verificarlo, VeriTracer, and VPREC.
  • Instrumented production-scale C/C++/Fortran scientific applications and parallel solvers; studied vectorization, communication costs, and precision–performance trade-offs.

Industry and Software Engineering Experience

Software Engineer, Numerical Computing Team

Hillsboro, Oregon, USA • –
Intel Corporation
  • Modernized functions in the Intel Mathematical Library and optimized approximately half of the targeted routines; implemented continuous-integration validation to ensure numerical library quality.

Software Developer

Versailles, France • –
Université de Versailles Saint-Quentin-en-Yvelines
  • Developed a multithreaded capture-and-replay mechanism in C for CERE, doubling experiment scalability; microbenchmarked NAS kernels for energy–performance modelling.

Software Developer

Bruyères-le-Châtel, France • –
Exascale Computing Research
  • Implemented value profiling with LLVM/Python and automatic specialization of C functions; evaluated the resulting speedups.

Peer-Reviewed Journal Articles

  1. I. Gonzalez Pepe, H. Akhaddar, T. Glatard, Y. Chatelain. “Fuzzy PyTorch: Rapid Numerical Variability Evaluation for Deep Learning Models.” Transactions on Machine Learning Research, 2026. [pdf]
  2. Y. Chatelain, A. Sokolowski, M. Sharp, J.-B. Poline, T. Glatard. “The practical impact of numerical variability on structural MRI measures of Parkinson's disease.” Scientific Reports, 2026. [doi]
  3. J. Sanz-Robinson, M. Wang, B. McPherson, Y. Chatelain, D. Kennedy, T. Glatard, J.-B. Poline. “Open-source platforms to investigate analytical flexibility in neuroimaging.” Imaging Neuroscience, 2025. [pdf]
  4. M. Dugré, Y. Chatelain, T. Glatard. “An Analysis of Performance Bottlenecks in MRI Pre-Processing.” GigaScience 14, 2025. [pdf]
  5. Y. Chatelain, L. Tetrel, C. J. Markiewicz, M. Goncalves, G. Kiar, O. Esteban, P. Bellec, T. Glatard. “A numerical variability approach to results stability tests and its application to neuroimaging.” IEEE Transactions on Computers 74(1), 2024. [pdf]
  6. A. Sokolowski, N. Bhagwat, Y. Chatelain, M. Dugre, A. Hanganu, O. Monchi, B. McPherson, M. Wang, J.B. Poline, M. Sharp, T. Glatard. “Longitudinal brain structure changes in Parkinson's disease: a replication study.” PLOS ONE, 2024. [pdf]
  7. I. Gonzalez Pepe, Y. Chatelain, G. Kiar, T. Glatard. “Numerical Stability of DeepGOPlus Inference.” PLOS ONE, 2024. [pdf]
  8. Y. Chatelain, N. Yong, G. Kiar, T. Glatard. “PyTracer: Automatically profiling numerical instabilities in Python.” IEEE Transactions on Computers, 2022. [pdf]
  9. G. Kiar, Y. Chatelain, A. Salari, A. C. Evans, T. Glatard. “Data Augmentation Through Monte Carlo Arithmetic Leads to More Generalizable Classification in Connectomics.” Neurons, Behavior, Data Analysis and Theory, 2021. [pdf]
  10. G. Kiar, Y. Chatelain, P. de Oliveira Castro, E. Petit, A. Rokem, G. Varoquaux, B. Misic, A. C. Evans, T. Glatard. “Numerical Uncertainty in Analytical Pipelines Leads to Impactful Variability in Brain Networks.” PLOS ONE, 2021. [pdf]
  11. M. Popov, C. Akel, Y. Chatelain, W. Jalby, P. de Oliveira Castro. “Piecewise holistic autotuning of parallel programs with CERE.” Concurrency and Computation: Practice and Experience 29, 2017. [pdf]

Peer-Reviewed Conference and Workshop Papers

  1. I. Gonzalez Pepe, V. Sivakolunthu, Y. Chatelain, T. Glatard. “Uncertain but Useful: Leveraging CNN Training Variability into Data Augmentation.” UNSURE, MICCAI, 2026 (poster). [paper]
  2. N. Mirhakimi, Y. Chatelain, J.-B. Poline, T. Glatard. “Numerical Uncertainty in Linear Registration: An Experimental Study.” UNSURE, MICCAI, 2025. [pdf]
  3. G. Vila, E. Medernach, I. Gonzalez Pepe, A. Bonnet, Y. Chatelain, M. Sdika, T. Glatard, and S. Camarasu-Pop. “The Impact of Hardware Variability on Applications Packaged with Docker and Guix: a Case Study in Neuroimaging.” ACM REP, 2024. Best Paper Award [pdf]
  4. I. Gonzalez Pepe, V. Sivakolunthu, H. Lang Park, Y. Chatelain, T. Glatard. “Numerical Uncertainty of Convolutional Neural Networks Inference for Structural Brain MRI Analysis.” UNSURE, MICCAI, 2023. [pdf]
  5. M. Des Ligneris, A. Bonnet, Y. Chatelain, T. Glatard, M. Sdika, G. Vila, V. Wargnier-Dauchelle, S. Pop, C. Frindel. “Reproducibility of tumor segmentation outcomes with a deep learning model.” IEEE ISBI, 2023. [pdf]
  6. M. Vicuna, M. Khannouz, G. Kiar, Y. Chatelain, T. Glatard. “Reducing numerical precision preserves classification accuracy in Mondrian Forests.” IEEE Big Data, 2021. [pdf]
  7. A. Salari, Y. Chatelain, G. Kiar, T. Glatard. “Accurate simulation of operating system updates in neuroimaging using Monte-Carlo arithmetic.” UNSURE, MICCAI, 2021. [pdf]
  8. Y. Chatelain, E. Petit, P. de Oliveira Castro, G. Lartigue, D. Defour. “Automatic exploration of reduced floating-point representations in iterative methods.” Euro-Par, 2019. [pdf]
  9. Y. Chatelain, P. de Oliveira Castro, E. Petit, D. Defour, J. Bieder, M. Torrent. “VeriTracer: Context-enriched tracer for floating-point arithmetic analysis.” IEEE ARITH, 2018. [pdf]

Selected Preprints and Work Under Review

  1. Y. Chatelain, P. de Oliveira Castro. “Stochastic Rounding in Low-Precision Transformer Inference: A Variable-Precision Emulation Study of a Small GPT-2.” arXiv:2610.01889, 2026. [pdf]
  2. A. Encin, I. Gonzalez Pepe, Y. Chatelain, E. Dickie, T. Glatard. “Untrained Convolutional Neural Networks as Feature Extractors for Structural MRI.” bioRxiv, 2026. [pdf]
  3. I. Gonzalez Pepe, V. Sivakolunthu, J. Fortin, Y. Chatelain, T. Glatard. “Conservative & Aggressive NaNs Accelerate U-Nets for Neuroimaging.” arXiv:2601.17180, 2026. [pdf]
  4. M. Alizadeh, Y. Chatelain, G. Kiar, T. Glatard. “Numerical Variability of Functional MRI Graph Measures.” bioRxiv, 2025. [pdf]

Selected Scientific Presentations

  • ATPESC 2026 (Argonne National Laboratory, St. Charles, IL) — invited lecture and hands-on tutorial “Verificarlo: Debugging and optimising floating-point calculations”, Mixed Precision Computing track.
  • OHBM 2022, Glasgow — result-stability testing and the long-term reproducibility of fMRIPrep.
  • SciPy 2021 — Fuzzy environments for perturbing, evaluating, and leveraging numerical uncertainty in the scientific Python ecosystem.
  • IXPUG 2018 (Intel, Hillsboro) and IXPUG 2019 (CERN, Geneva) — HPC numerical computing and software tooling.
  • APS 2018 — challenges in scaling ABINIT electronic-structure simulations to exascale.

Software Contributions

Verificarlo / Interflop LLVM instrumentation for C/C++/Fortran programs and reusable arithmetic backends for Monte Carlo arithmetic, reduced precision, and vectorized execution across architectures.
PRISM / Fuzzy PyTorch Creator of PRISM. Vectorized probabilistic rounding for low-overhead analysis of numerical variability in optimized PyTorch models; CPU foundation for future GPU and accelerator instrumentation.
PyTracer / VeriTracer Context-aware profiling and localization of numerical instabilities in Python and compiled scientific applications.
VPREC Dynamic simulation and exploration of reduced floating-point formats; applied to parallel fluid-mechanics solvers and precision optimization for neuroimaging registration.
Stability methodology Result-level acceptability tests and propagation of numerical variability to effect sizes, correlations, longitudinal models, and group-level inferences.

Funding and Awards

Concordia Horizon Postdoctoral Fellowship

CAD $50,000/year • –
Principal investigator • Numerical instabilities in neuroimaging

AccelNet IN-BIC Research Grant

US $10,000 •
Principal investigator • Numerical stability of brain tractometry with PyAFQ

Best Paper Award, ACM REP

Co-author •
2nd ACM Conference on Reproducibility and Replicability

Teaching Experience

Invited Lecturer — ATPESC 2026

St. Charles, IL, USA •
Argonne National Laboratory • Argonne Training Program on Extreme-Scale Computing
  • Lecture and hands-on tutorial (Google Colab) on Verificarlo in the Mixed Precision Computing track: debugging and optimizing floating-point computations, stochastic arithmetic, reduced precision (VPREC), and instrumentation of HPC and deep-learning codes.

Course Instructor — Programming Foundations

Montréal, Canada •
Concordia University • first-year undergraduate course
  • Introduced programming, classes, and objects in C++ through worked examples, regular practice, testing, and conceptual understanding.

Teaching Assistant / Lecturer — Compilers

Versailles, France • –
Université Paris-Saclay • third-year undergraduate course
  • Taught the compilation pipeline, from parsing to assembly generation; supervised the implementation of a functional mini-compiler for Tiger.

Teaching Assistant — Advanced Algorithms

Versailles, France • –
Université Paris-Saclay • third-year undergraduate course
  • Led tutorials on time and space complexity, recursion, divide-and-conquer methods, and graph algorithms; delivered an introductory lecture.

Teaching Assistant — Parallel Architectures

Versailles, France • –
Université Paris-Saclay • first-year master's course
  • Led tutorials on shared- and distributed-memory programming, cache coherence, and interconnection networks; assessed paper reviews and oral presentations.

Courses prepared to teach: parallel systems and HPC (OpenMP, MPI, CUDA); operating systems and software systems; compilers; numerical computing and scientific software reliability; algorithms; and introductory programming. Able to teach in English or French.

Research Mentorship

These mentorships were conducted under my supervisor's formal responsibility, while I played an active role in day-to-day supervision, including defining research topics, monitoring scientific progress, and providing methodological guidance.

Level Student Topic Period
PhD Mina Alizadeh Numerical stability of functional neuroimaging –present
PhD Inés Gonzalez Pepe Numerical stability of deep learning in bioinformatics –present
PhD Mathieu Dugré Reduced precision for neuroimaging applications –
PhD Ali Salari Computing environments for neuroimaging pipelines –
Master's Inés Gonzalez Pepe Numerical stability of DeepGOPlus inference (50% mentorship) –
Master's Damien Thénot Java environment for VeriTracer (50% supervision)
Undergraduate Jacob Fortin U-Net acceleration through NaN pooling/unpooling (50%)
Undergraduate Nigel Yong PyTracer performance optimization (50%)
Undergraduate Marc Vicuna Reduced precision in Mondrian forests (co-supervision)

Education

PhD in Computer Science

Versailles, France •
Université Paris-Saclay / UVSQ
Dissertation: Debugging and optimization tools for floating-point computations in HPC.

Master's in High-Performance Computing and Numerical Simulation

Versailles, France •
Université Paris-Saclay / UVSQ

Bachelor's in Computer Science

Orsay, France •
Université Paris-Saclay

Technical Expertise

HPC OpenMP, MPI, multithreading and multiprocessing, Slurm, distributed experiments, profiling, strong and weak scaling, and CPU/GPU and heterogeneous systems.
Compilers LLVM/Clang, compiler passes, binary instrumentation, Intel/ARM ISAs, assembly, vectorization, and optimized numerical kernels.
Languages and platforms C, C++, Fortran, Python, and OCaml; Linux, Docker, Guix, continuous integration and deployment, and reproducible environments and workflows.
Numerical computing and AI Floating-point, Monte Carlo and stochastic arithmetic, mixed/reduced precision, PyTorch, NumPy/SciPy, FreeSurfer, ANTs, PyAFQ.

Professional Profile

Languages French (native); English (professional proficiency).
Open science Open-source software, reproducible containers, automated stability reports, and continuous integration.
Professional status Prepared to complete the requirements for membership in the Ordre des ingénieurs du Québec.