Mode Collapse
Mila Quebec AI Institute. Humanistic technology bretheren.
Joseph David Viviano
6666 Rue Saint Urbain
Montreal, Quebec
Canada, H2S 3H1
I believe that the discoveries and technology we need to address our era’s most pressing issues cannot be designed without AI first giving us better eyes to see. My main interests lie in the design of AI for Science systems, and leveraging them to enable scientific breakthroughs. Science stands to benefit more from the tidal wave of artificial intelligence currently permeating society, and I want to build multi-agent collaborative systems to amplify the speed at which human research teams can make new discoveries.
I first studied Psychology at Queen’s University, then Neuroscience at York University. I worked on psychiatric biomarker development at CAMH before studying Machine Learning at Mila and the Université de Montréal. I did internships on ML applications in Radiology with Imagia and uncertainty estimation with the Google search ad team. I worked on portfolio construction for CDPQ as part of my work with the Applied Machine Learning team at Mila, and worked on deep learning method for RNA biology with Deep Genomics, before my current contract Research Engineer position with Yoshua Bengio’s group at Mila. I’m now focused on open source tool development and applications of machine learning, particularly GFlowNets, in the area of biological discovery. Some of my recent work includes the development of an antibody loop sampler in collaboration with Amgen and researching asynchronous distributed training protocols for GFlowNets in collaboration with Intel.
I consider myself a strong generalist who loves both science and engineering, who has a knack for building bridges between disciplines and between scientific knowledge and useful tools.
I’m most interested in use of AI-powered laboratories to power scientific breakthroughs. My primary interest resides in scientific fields where experimental protocols are aligned with agentic lab in the loop workflows. Long term, I hope to leverage such systems to discover new technologies to improve the sustainability of our planet and societies, improve food security, the lengthen healthy human lifespan.
I’m also an amateur photographer and musician. Head over to the creative section to see some samples of my stuff.

news
| Oct 20, 2024 | Very honoured to give the invited keynote for the FACSS SCIX 2024 conference, “Can an AI Understand a Scientist?” |
|---|---|
| Mar 01, 2023 | Excited to be joining Yoshua Bengio’s team to assist with the creation of open source tools for GFlowNets, and other AI for Science applications, particularly in Biology! |
selected publications
- torchgfn: A PyTorch GFlowNet libraryarXiv preprint arXiv:2305.14594, 2025
- Action abstractions for amortized samplingIn International Conference on Learning Representations (ICLR), 2025
- TorchXRayVision: A library of chest X-ray datasets and modelsIn International Conference on Medical Imaging with Deep Learning, 2022
- What’s in the Box? A Preliminary Analysis of Undesirable Content in the Common Crawl CorpusIn ACL-IJCNLP, 2021
- Saliency is a Possible Red Herring When Diagnosing Poor GeneralizationIn International Conference on Learning Representations (ICLR), 2021
- Problems in the deployment of machine-learned models in health careCmaj, 2021
- Resting-state connectivity biomarkers of cognitive performance and social function in individuals with schizophrenia spectrum disorder and healthy control subjectsBiological Psychiatry, 2018
- Modeling and prediction of clinical symptom trajectories in Alzheimer’s disease using longitudinal dataPLoS computational biology, 2018