Home // PESARO 2019, The Ninth International Conference on Performance, Safety and Robustness in Complex Systems and Applications // View article

A GRU-based Meta-learning Model Based on Active Learning

Authors:
Honglan Huang
Shixuan Liu
Yanghe Feng
Jincai Huang
Zhong Liu

Keywords: active learning; meta learning; reinforcement learning; GRU.

Abstract:
In the realities of machine learning, labeling a data set may be expensive, tedious, or extremely difficult and it is often not easy to choose a common criteria for active learning to select samples for different data sets. In order to solve these difficulties, this paper introduces a GRU-based meta-learner model, which combines active learning with reinforcement learning and uses it in a stream-based one-shot learning task. Based on the uncertainty of the instances, the model learns an action strategy that determines when to predict or request the label of each instance. Through the experiments on Omniglot dataset, the model shows its ability to achieve a good prediction accuracy with few label requests.

Pages: 25 to 27

Copyright: Copyright (c) IARIA, 2019

Publication date: March 24, 2019

Published in: conference

ISSN: 2308-3700

ISBN: 978-1-61208-698-9

Location: Valencia, Spain

Dates: from March 24, 2019 to March 28, 2019