Computer Vision News - October 2020

AI application development platform 20 really looking forward to exploring this again from a commercial perspective in our future roadmap.” Healthcare use cases are a motivating example of why Snorkel needs to be easy to use for people who are domain rather than machine learning experts. Something Braden and Paroma have noticed when working on video-related computer vision work is a need for a layer of abstraction to know how to pull out the interpretable building blocks from the data. With MRI images, when writing labelling functions or trying to introduce noisy rules, they are written over some aspect or feature of the image, such as the area of the heart that you are seeing in the MRI. It falls under the same paradigm as Snorkel – the idea that there are noisy rules, you de-noise them, get training data, and then train a downstream machine learning model. Except now this is just writing the rules over interpretable building blocks of images or videos. The team are still tapped into the academic community and have maintained their connections with Stanford. One of their founders, Chris Ré , continues to be a professor there, and there are many others who they regularly work and chat with. Snorkel’s CEO, Alex Ratner , is a professor at University of Washington , and several other employees and advisors are professors at Brown and Wisconsin .