Numerical Reliability and Scientific Computing
Research and scientific software
I am a scientific associate in the Krembil Centre for Neuroinformatics at the Centre for Addiction and Mental Health (CAMH), Canada. My research topics include computer arithmetic, high-performance computing, reproducibility, and neuroimaging. In addition, my work aims to democratize the use of stability analysis of scientific computing codes through automatic tools to improve numerical quality.
I develop tools for observing and quantifying floating-point variability, and study how that variability affects scientific software and neuroimaging results.
For publications and collaboration, see Research, About, and Contact.
News
-
Citeometer: Live Citation, Readership, and Download Badges for Academic Sites
I built Citeometer, a lightweight, zero-dependency Web Component (
<cite-o-meter>) to display live scholarly publication metrics—citations, recent readership, PDF downloads, and software/dataset usage—on static academic websites and GitHub Pages. -
Stochastic rounding in low-precision transformer inference: preprint
I am thrilled to announce the preprint “Stochastic Rounding in Low-Precision Transformer Inference: A Variable-Precision Emulation Study of a Small GPT-2“ is available on arXiv.
-
CNN training variability for data augmentation: accepted at UNSURE 2026
I am thrilled to announce that our paper “Uncertain but Useful: Leveraging CNN Training Variability into Data Augmentation“ has been accepted as a poster at the International Workshop on Uncertainty for Safe Utilization of Machine Learning in Medical Imaging (UNSURE), MICCAI 2026.
-
Invited lecture at ATPESC 2026: Verificarlo
I gave an invited lecture and hands-on tutorial, “Verificarlo: Debugging and optimising floating-point calculations”, at the Argonne Training Program on Extreme-Scale Computing (ATPESC 2026), in the Track 4b: Mixed Precision Computing session (Thursday, July 30, 2026, Q Center, St. Charles, IL).
-
Numerical variability in Parkinson’s structural MRI: Scientific Reports study
This study evaluates how numerical variability in structural MRI processing can affect group comparisons and clinical associations in Parkinson’s disease research. The original post announced acceptance on July 27, 2026. The Scientific Reports article was published online on August 13, 2026. At this revision, the publisher identifies the available text as an unedited early-access manuscript.
-
Untrained CNN feature extractors for structural MRI: preprint
I am thrilled to announce the preprint “Untrained Convolutional Neural Networks as Feature Extractors for Structural MRI“ is available on bioRxiv.
-
Fuzzy PyTorch: evaluating numerical variability in deep learning
Fuzzy PyTorch evaluates arithmetic variability in deep-learning models by combining instrumented PyTorch with the Verificarlo compiler and PRISM probabilistic rounding. The paper appeared in Transactions on Machine Learning Research in January 2026; this post retains its original announcement date of February 18.
-
Numerical variability in Parkinson’s structural MRI: bioRxiv preprint
I am thrilled to announce the preprint “The practical impact of numerical variability on structural MRI measures of Parkinson’s disease“ is available on bioRxiv.
-
Numerical variability of functional MRI graph measures: preprint
I am thrilled to announce the preprint “Numerical Variability of functional MRI Graph Measures“ is available on bioRxiv.
-
Numerical uncertainty in linear registration: UNSURE 2025
I am thrilled to announce that the paper titled “Numerical uncertainty in linear registration: An experimental study“ is available.
-
CNN variability as data augmentation: arXiv preprint
I am thrilled to announce the preprint “Uncertain but useful: Leveraging cnn variability into data augmentation“ is available on arXiv.
-
Open-source platforms for analytical flexibility in neuroimaging
I am thrilled to announce that the paper titled “Open-source platforms to investigate analytical flexibility in neuroimaging“ is published in Imaging Neuroscience.
-
Floacon: A Web-Based Floating-Point Converter and Explorer
Floacon, a new web-based tool designed to help understand and explore floating-point numbers. This project aims to provide an interactive way to visualize floating-point formats, and to experiment with custom formats.
-
Performance bottlenecks in MRI preprocessing: GigaScience paper
I am thrilled to announce that the paper titled “An analysis of performance bottlenecks in MRI preprocessing“ has been accepted at GigaScience 25.
-
Numerical variability for results stability tests in neuroimaging
This work uses a numerical reference distribution to test whether a candidate scientific pipeline result remains compatible with a reference implementation. The original post announced acceptance in October 2024. The journal article appears in IEEE Transactions on Computers, volume 74, issue 1, pages 200–209, January 2025, with DOI 10.1109/TC.2024.3475586.
-
Hardware variability in containerized neuroimaging: ACM REP 2024 Best Paper Award
I am thrilled to announce that the paper titled “The Impact of Hardware Variability on Applications Packaged with Docker and Guix: a Case Study in Neuroimaging“ has been awarded the Best Paper Award at ACM REP 24.
-
Visualizing numerical instability with significant digits
This post introduces a significant-bit view of repeated arithmetic executions. The accompanying significantdigits package implements statistical estimators for stochastic arithmetic. Such estimates describe agreement within a chosen experiment; an external accuracy reference is needed to distinguish stable errors from correct results.