Meta Pseudo Labels

This video explains Meta Pseudo Labels! This is a really interesting algorithm for dynamically adapting the ground truth targets (y) while training a classification network. This is done by training a teacher network via gradient through a gradient to label images such that the classifier performs well on a held-out validation set. Thanks for watching! Please Subscribe! Paper Links: Meta Pseudo Labels: https://ift.tt/2JkDP4U UDA: https://ift.tt/37B0FPb Generative Teaching Networks: https://ift.tt/2w8Yd64 When Does Label Smoothing Help? https://ift.tt/2JuThuN Self-Training with Noisy Student: https://ift.tt/2warw7S FixMatch: https://ift.tt/3bU1Bll RandAugment: https://ift.tt/2QRHZ8l Thanks for watching, please subscribe!