Minimax rate of robust estimation when different samples have different but known rates of corruption.

We introduce a novel metric for evaluating the quality of (potentially stochastic) optimal transport maps, which enables efficient and robust estimation in previously unexplored regimes of practical significance.