The author proposes an efficient method, ENOT, using expectile regularization to improve Neural Optimal Transport training by approximating conjugate potentials accurately and quickly.
