Jared Dunnmon

I am currently a postdoctoral researcher with Chris Ré at Stanford.

My research focuses on developing weakly supervised machine learning systems to support applications in areas such as medicine, energy and environment, and intelligence analysis wherein the costs of failure are high and labeled data is scarce.

By modeling noisy, programmatic sources of supervision applied to unlabeled data, directly integrating human domain knowledge, and leveraging related tasks, my work aims to enable reliable application of state-of-the-art machine learning models to these problems with the speed, scale, and performance levels required for practical deployment.

Email     LinkedIn     Google Scholar     GitHub    


  • [May 2020] Paper on cross-modal data programming for medical imaging published in Cell Patterns.
  • [April 2020] Exciting collaboration with neurologists on weak supervision for EEG monitoring reported in NPJ Digital Medicine.
  • [February 2020] Recent work using weak supervision for science and medicine featured in MLSys 2020 keynote; check out our blog post for more!
  • [February 2020] Research describing hidden stratification as a problem in medical imaging to be featured as an oral spotlight in ACM CHIL 2020.
  • [August 2019] Recent paper on using multiple levels of weak supervision to train high-performance hemorrhage detection models will be presented as an oral spotlight at MICCAI 2019.
  • [April 2019] Presented work on using eye tracking to improve computer vision algorithms at ICLR Limited Labeled Data workshop.
  • [Nov. 2018] Clinical journal Radiology features our work on automated radiograph triage using convolutional neural networks; check out both the article and the associated editorial.
  • [Oct. 2018] Paper describing how we can model multi-task weak supervision using a matrix completion-style approach accepted to AAAI 2019.
  • [May 2018] "Snorkel MeTaL: Weak Supervision for Multi-Task Learning" presented as a long talk at SIGMOD DEEM Workshop.
  • [Dec. 2017] Presented our work on using human domain knowledge for better data augmentation at NeurIPS 2017; check out the paper and blog post.


Snorkel is a system for rapidly creating, modeling, and managing training data by leveraging a variety of weak supervision sources in a principled manner. Today's state-of-the-art machine learning models require massive labeled training sets, which usually do not exist for real-world applications.

Particularly complex problems are often composed of multiple tasks, and may have many different types of weak supervision that provide labels for one or more of these tasks. In Snorkel MeTaL, we use a new modeling approach to denoise this massively multi-task weak supervision before training an auto-compiled multi-task neural network.

Preprints and Workshops
Improving Sample Complexity with Observational Supervision
Khaled Saab*, Jared Dunnmon*, Alexander Ratner, Daniel Rubin, and Christopher Ré
ICLR Limited Labeled Data Workshop, 2019.
Cross-Modal Data Programming Enables Rapid Medical Machine Learning
Jared Dunnmon*, Alexander Ratner*, Nishith Khandwala, Khaled Saab, Matthew Markert, Hersh Sagreiya, Roger Goldman, Christopher Lee-Messer, Matthew Lungren, Daniel Rubin, and Christopher Ré
Patterns, 2020.
Weak Supervision as an Efficient Approach for Automated Seizure detection in Electroencephalography
Jared Dunnmon*, Khaled Saab*, Christopher Ré, Daniel Rubin, and Christopher Lee-Messer
NPJ Digital Medicine, 2020.
Hidden Stratification Causes Clinically Meaningful Failures in Machine Learning for Medical Imaging
Luke Oakden-Rayner*, Jared Dunnmon*, Gustavo Carneiro, and Christopher Ré
ACM Conference on Health, Inference, and Learning, 2020 (Oral Spotlight).
Doubly Weak Supervision of Deep Learning Models for Head CT
Khaled Saab*, Jared Dunnmon*, Roger Goldman, Hersh Sagreiya, Alexander Ratner, Christopher Ré, and Daniel Rubin
MICCAI, 2019 (Oral Spotlight).
Weakly Supervised Classification of Rare Aortic Valve Malformations Using Unlabeled Cardiac MRI Sequences
Jason Fries, Paroma Varma, Vincent Chen, Ke Xiao, Heliodoro Tejada, Saha Priyanka, Jared Dunnmon, Henry Chubb, Shiraz Maskatia, Madalina Fiterau, Scott Delp, Euan Ashley, Christopher Ré, and James Priest
Nature Communications, 2019.
Training Complex Models with Multi-Task Weak Supervision
Alexander Ratner, Braden Hancock, Jared Dunnmon, Frederic Sala, Shreyash Pandey, and Christopher Ré
Proceedings of the 33rd AAAI Conference on Artificial Intelligence, 2019.
Assessment of Convolutional Neural Networks for Automated Triage of Chest Radiographs
Jared Dunnmon, Darvin Yi, Curtis Langlotz, Christopher Ré, Daniel Rubin, and Matthew Lungren
Radiology, 2018.
[Supplement, Editorial]
Snorkel MeTaL: Weak Supervision for Multi-Task Learning
Alexander Ratner, Braden Hancock, Jared Dunnmon, Shreyash Pandey, and Christopher Ré
Proceedings of the Second ACM DEEM Workshop, 2018.
Transformational Adversarial Networks for Data Augmentation
Alexander Ratner, Henry Ehrenberg, Zeshhan Hussain, Jared Dunnmon, and Christopher Ré
Advances in Neural Information Processing Systems, 2017.
An Investigation of Internal Flame Structure in Porous Media Combustion via X-ray Computed Tomography
Jared Dunnmon, Sadaf Sobhani, Meng Wu, Rebecca Fahrig, and Matthias Ihme
Proceedings of the Combustion Institute, 36(3):4399-4408, 2017.
Investigation of Lean Combustion Stability and Pressure Drop in Porous Media Burners
Sadaf Sobhani, Bret Haley, David Bartz, Jared Dunnmon, Jonathan Sullivan, and Matthias Ihme
Proceedings of Turbo Expo, 2017.
Characterization of Scalar Mixing in Dense Gaseous Jets Using X-Ray Computed Tomography
Jared Dunnmon, Sadaf Sobhani, Tae Wook Kim, Anthony Kovscek, and Matthias Ihme
Experiments in Fluids, 56(10):193, 2015.
2012 2011
Power Extraction from Aeroelastic Limit Cycle Oscillations
Jared Dunnmon, Samuel Stanton, Brian Mann, and Earl Dowell
Journal of Fluids and Structures, 27(8):1182-1198, 2011.
Other Collaborations

( * Equal Contributors)