Can I read Model-based Meta Learning on EtoBox?

Model-based Meta Learning by Pengyu Yuan; Hien Van Nguyen is a book available to read on EtoBox.

What is Model-based Meta Learning about?

The Neural Turing Machine (NTM) is made up of a controller, such as a feed-forward network or an LSTM, that communicates with an external memory module through a number of read and write heads . Memory encoding and retrieval in an NTM external memory module is fast. These operations insert or retrieve vector representations out of memory at every time step. NTM is capable of both long-term storage via gradual weight updates and short-term storage via its external memory module. This capability makes the NTM an ideal option for meta learning. As a result, if an NTM can learn a general strategy to determine what to store in the memory and how to utilize them for predictions, it may be able to generate correct predictions for tasks with limited training samples.In our model, the controllers are either LSTMs or feed-forward networks. The controller communicates with an external memory module through read and write heads, which respectively extract representations from memory and insert representations into memory. Given some input, x t , the controller generates a key, k t , based on the controller's hidden states:

Author

Pengyu Yuan; Hien Van Nguyen

Publisher

Elsevier

Published

2023

Language

EN