Tweeted By @svlevine

In the real world, the test distribution never actually matches the training distribution. Adaptive risk minimization (ARM) addresses distributional shift by adapting to it, without labels -- just from seeing a group of test inputs rather than a single individual test input. https://t.co/BmB3xjMazn pic.twitter.com/vOrVzw4pDX

— Sergey Levine (@svlevine) July 7, 2020