Few-Shot Learning with Graph Neural Networks
This work addresses few-shot learning for machine learning applications, offering an incremental improvement by unifying and extending existing approaches.
The authors tackled few-shot learning by framing it as inference on a partially observed graphical model, using a graph neural network architecture that generalizes existing models and shows improved numerical performance, with extensions to semi-supervised and active learning.
We propose to study the problem of few-shot learning with the prism of inference on a partially observed graphical model, constructed from a collection of input images whose label can be either observed or not. By assimilating generic message-passing inference algorithms with their neural-network counterparts, we define a graph neural network architecture that generalizes several of the recently proposed few-shot learning models. Besides providing improved numerical performance, our framework is easily extended to variants of few-shot learning, such as semi-supervised or active learning, demonstrating the ability of graph-based models to operate well on 'relational' tasks.