Training Neural Networks with Fixed Sparse Masks

A Twitter thread by one of the authors, for more context: https://twitter.com/colinraffel/status/1468618801134592012