view reply Great work, super interesting article! You mention that the best curriculum is an annealing strategy. Is that between two or multiple datasets and did you also test annealing between different mixtures? Thanks!