Home // DBKDA 2021, The Thirteenth International Conference on Advances in Databases, Knowledge, and Data Applications // View article

A Framework for Improving Offline Learning Models with Online Data

Authors:
Sabrina Luftensteiner
Michael Zwick

Keywords: Online Learning; Catastrophic Forgetting; Regression; Domain Adaption

Abstract:
The usage of available online data is rising as machines get equipped with more sensors to control and monitor processes. The produced data can be used to directly fit existing prediction models to enhance their accuracy, adapt to alterations within the environment and avoid the training of new models. During the online learning step, which is used for the adaptation of the models using online data, catastrophic forgetting of already learned tasks may occur. We propose a new framework that utilizes several state-of-the-art methods in deep learning, as well as machine learning to minimize catastrophic forgetting. The methods range from memory-based approaches to methods for loss calculation and different optimizers, whereat the framework also provides possibilities to compare the methods and their impact with each other. The proposed framework is specifically tailored for regression problems, focusing on industrial settings in the experiments section. It is able to cope with single and multi-task models, is expandable and enables a high variety of configuration possibilities for adaptation to a given problem.

Pages: 32 to 37

Copyright: Copyright (c) IARIA, 2021

Publication date: May 30, 2021

Published in: conference

ISSN: 2308-4332

ISBN: 978-1-61208-857-0

Location: Valencia, Spain

Dates: from May 30, 2021 to June 3, 2021