ACDC: online unsupervised cross-domain adaptation
Files
(Published version)
Date
2022
Authors
De Carvalho, M.
Pratama, M.
Zhang, J.
Yapp Kien Yee, E.
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Journal article
Citation
Knowledge-Based Systems, 2022; 253(253, article no.109486):1-15
Statement of Responsibility
Conference Name
Abstract
We consider the problem of online unsupervised cross-domain adaptation, where two independent but related data streams with different feature spaces – a fully labeled source stream and an unlabeled target stream – are learned together. Unique characteristics and challenges such as covariate shift, asynchronous concept drifts, and contrasting data throughput arise. We propose ACDC, an adversarial unsupervised domain adaptation framework that handles multiple data streams with a complete self-evolving neural network structure that reacts to these defiances. ACDC encapsulates three modules into a single model: A denoising autoencoder that extracts features, an adversarial module that performs domain conversion, and an estimator that learns the source stream and predicts the target stream. ACDC is a flexible and expandable framework with little hyper-parameter tunability. Our experimental results under the prequential test-then-train protocol indicate an improvement in target accuracy over the baseline methods, achieving more than a 10% increase in some cases.
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
Copyright 2022 Elsevier
Access Condition Notes: Accepted manuscript available after 1 October 2024