1 paper · 1 filter
Peter Dixon, A. Pavan, Jason Vander Woude +1
We investigate replicable learning algorithms. Ideally, we would like to design algorithms that output the same canonical model over multiple runs, even when different runs observe…