Joseph Paul Cohen PhD

Deep Learning for Ultrasound, Butterfly Network
Former Postdoctoral Research Fellow, AIMI, Stanford University
Former Postdoctoral Fellow, Mila, Quebec AI Institute
Director, The Institute for Reproducible Research

My goal is to democratize access to healthcare to provide the highest quality healthcare to everyone (specifically those not served by the current system; 8.3% in the US and 25% globally). Automation and AI can increase the supply of providers to fill this need. I am working to identify and overcome issues limiting the deployment of AI tools in healthcare. My core research directions are representation learning, generalization, and out of distribution detection.

My interests are in medical applications of deep learning:

  • Medical Imaging: radiology, histology, microscopy, cell counting
  • Genomics: gene representations, scRNA-Seq, cancer subtype/phenotype prediction
  • Clinical: survival/event prediction, automated triage

As well as core deep learning:

  • Generalization: unsupervised representation learning, meta-learning, concept representations
  • Uncertainty: out of distribution detection, model calibration
  • Attribution: prediction explanation, model interpretation

Bio: Joseph Paul Cohen is a researcher and pragmatic engineer. He currently focuses on the challenges in deploying AI tools in medicine specifically computer vision and genomics. He maintains many open source projects including Chester the AI radiology assistant, TorchXRayVision, and BlindTool – a mobile vision aid app. He is the director of the Institute for Reproducible Research, a US non-profit which operates and Academic Torrents.

, LinkedIn, Google Scholar,

Selected Projects