You are on page 1of 1

This file is to run sample model in the machine learning.

-tasks are implicitly related--in previous work

-task are explicitly related--in this work

- graph based meta-learner (called GPN [Gated Propagation Network]) for prototype

based classification in which prototype

is generated for each class.

-In GPN, attention mechanism aggregates messages from neighboring classes with a gate choosing
between aggregated message and the message from the class itself.

- many-shot to few-shot: generated from subgaph sampling

-graph with each node as a class and each edges as a children or parent class.

-Goal: learn meta learner for each task M(T,theta) where T is each task and theta is the model
parameter

previous GNN:

-uses each examples as a node while this method used prototypical representation of a class as a node.

You might also like