Toward Multimodal Model-Agnostic Meta-Learning

Risto Vuorio
Risto Vuorio
Hexiang Hu
Hexiang Hu

arXiv: Learning, Volume abs/1812.07172, 2018.

Cited by: 0|Bibtex|Views26|Links
EI

Abstract:

Gradient-based meta-learners such as MAML are able to learn a meta-prior from similar tasks to adapt to novel tasks from the same distribution with few gradient updates. One important limitation of such frameworks is that they seek a common initialization shared across the entire task distribution, substantially limiting the diversity of ...More

Code:

Data:

Your rating :
0

 

Tags
Comments