Good introductory article about Transfer Learning byMachine Learning Mastery. Kerasis quite flexible for this task. It would be nice if an end-to-end example can be shown on how to train a model using transfer learning. Another approach which fast.ai library uses is to first freeze the VGG layers and just train the last couple of Dense layers so those are quickly trained. Then ‘unfreeze’ all layers and do some fine-tuning training epochs.
Written on December 30, 2020 by Michael Li.
Originally published on Medium