Few-Shot Learning with Reptile - Keras Code Examples

This video walks through an implementation of Reptile in Keras using the Omniglot dataset. I was really inspired by this example, I think the Omniglot challenge of dynamically being able to recombine characters to form new alphabets is an incredibly interesting problem, connecting Human and Artificial Intelligence. I hope you found this example interesting as well, please check out the rest of the Keras Code Example playlist! Content Links: Few-shot learning with reptile: https://keras.io/examples/vision/reptile/ On First-Order Meta Learning: https://arxiv.org/pdf/1803.02999.pdf MAML: https://arxiv.org/pdf/1703.03400.pdf Generative Teaching Networks: https://arxiv.org/pdf/1912.07768.pdf Teaching with Commentaries: https://arxiv.org/pdf/2011.03037.pdf Meta Pseudo Labels: https://arxiv.org/pdf/2003.10580.pdf Reptile Chapters 0:00 Beginning 0:44 Reptile Motivation 5:05 Hyperparameters of Meta-Training 5:48 DataLoader 12:04 Visualizing Omniglot Samples 13:50 The Model 14:48 The Training Loop 19:52 Results