Takeaways from our robust injury classifier project [Redwood Research]

Abstract

With the benefit of hindsight, we have a better sense of our takeaways from our first adversarial training project (paper). Our original aim was to use adversarial training to make a system that (as…