[Few-shot learning][2.5] Beyond MAML with MAML++ скачать в хорошем качестве

[Few-shot learning][2.5] Beyond MAML with MAML++

In this episode I provide an overview of the instability problems in Model-Agnostic Meta-Learning (MAML). In particular, I give an overview of MAML++ which has been introduced in the paper "How to train your MAML" (2019, ICLR). MAML++ solve the instability issues of MAML by using a multi-step loss and by (meta)learning a separate set of learning rates. My Blog: https://mpatacchiola.github.io/blog/ GitHub: https://github.com/mpatacchiola Linkedin:   / massimiliano-patacchiola-94579b140   #machinelearning #deeplearning #reinforcementlearning #robotics #artificialintelligence #ai #neuralnetworks