Domain adaptation from multiple sources via auxiliary classifiers

Duan, Lixin; Tsang, Ivor W.; Xu, Dong; Chua, Tat-Seng

doi:10.1145/1553374.1553411

Cited by 280 publications

(226 citation statements)

References 11 publications

Supporting

Mentioning

222

Contrasting

Order By: Relevance

“…In particular, amongst TL baselines we chose: No transfer: Regularized Least Squares (RLS) algorithm trained solely on the target data; Best source: indicates the performance of the best source classifier selected by its score on the testing set. This is a pseudo-indicator of what an HTL can achieve; AverageKT: obtained by averaging the predictions of all the source classifiers; RLS src+feat: RLS trained on the concatenation of feature descriptors and source classifier predictions; MultiKT · 2 : HTL algorithm by [24] selecting β in (1) by minimizing the leave-one-out error subject to β 2 ≤ τ ; MultiKT · 1 : similar to previous, but applying the constraint β 1 ≤ τ ; DAM: An HTL algorithm by [9], that can handle selection from multiple source hypotheses. It was shown to perform better than a well known and similar ASVM [26] algorithm.…”

Section: Methodsmentioning

confidence: 99%

Transfer Learning Through Greedy Subset Selection

Kuzborskij

Orabona

Caputo

2015

Image Analysis and Processing — ICIAP 2015

View full text Add to dashboard Cite

Abstract. We study the binary transfer learning problem, focusing on how to select sources from a large pool and how to combine them to yield a good performance on a target task. In particular, we consider the transfer learning setting where one does not have direct access to the source data, but rather employs the source hypotheses trained from them. Building on the literature on the best subset selection problem, we propose an efficient algorithm that selects relevant source hypotheses and feature dimensions simultaneously. On three computer vision datasets we achieve state-of-the-art results, substantially outperforming transfer learning and popular feature selection baselines in a small-sample setting. Also, we theoretically prove that, under reasonable assumptions on the source hypotheses, our algorithm can learn effectively from few examples.

show abstract

Section: Methodsmentioning

confidence: 99%

Transfer Learning Through Greedy Subset Selection

Kuzborskij

Orabona

Caputo

2015

Image Analysis and Processing — ICIAP 2015

View full text Add to dashboard Cite

show abstract

“…Model-based domain adaptation approaches [22,23] discover an adaptive classifier that performs well on the target data. In these models, the learned classifier transfers model parameters from the source to the target domain without any change in the feature space.…”

Section: Related Workmentioning

confidence: 99%

Exploiting kernel-based feature weighting and instance clustering to transfer knowledge across domains

Tahmoresnezhad¹,

Hashemi²

2017

Turk J Elec Eng & Comp Sci

View full text Add to dashboard Cite

Learning invariant features across domains is of vital importance to unsupervised domain adaptation, whereclassifiers trained on the training examples (source domain) need to adapt to a different set of test examples (target domain) in which no labeled examples are available. In this paper, we propose a novel approach to find the invariant features in the original space and transfer the knowledge across domains. We extract invariant features of input data by a kernel-based feature weighting approach, which exploits distribution difference and instance clustering to find desired features. The proposed method is called the kernel-based feature weighting (KFW) approach and benefits from the maximum mean discrepancy to measure the difference between domains. KFW uses condensed clusters in the reduced domains, the domains that do not contain variant features, to enhance the classification performance. Simultaneous use of feature weighting and instance clustering increases the adaptation and classification performance. Our approach automatically discovers the invariant features across domains and employs them to bridge between source and target domains. We demonstrate the effectiveness of our approach in the task of artificial and real world dataset examinations.Empirical results show that the proposed method outperforms other state-of-the-art methods on the standard transfer learning benchmark datasets.

show abstract

“…While several existing single source adaptation techniques can be extended to multi-source adaptation, the literature in multi-source adaptation can be broadly categorized as: 1) feature representation approaches (Chattopadhyay et al, 2012;Sun et al, 2011;Duan et al, 2009;Duan et al, 2012; …”

Section: Related Workmentioning

confidence: 99%

Cross-domain Text Classification with Multiple Domains and Disparate Label Sets

Bhatt

Sinha

Roy

2016

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

View full text Add to dashboard Cite

Advances in transfer learning have let go the limitations of traditional supervised machine learning algorithms for being dependent on annotated training data for training new models for every new domain. However, several applications encounter scenarios where models need to transfer/adapt across domains when the label sets vary both in terms of count of labels as well as their connotations. This paper presents first-of-its-kind transfer learning algorithm for cross-domain classification with multiple source domains and disparate label sets. It starts with identifying transferable knowledge from across multiple domains that can be useful for learning the target domain task. This knowledge in the form of selective labeled instances from different domains is congregated to form an auxiliary training set which is used for learning the target domain task. Experimental results validate the efficacy of the proposed algorithm against strong baselines on a real world social media and the 20 Newsgroups datasets.

show abstract

Domain adaptation from multiple sources via auxiliary classifiers

Cited by 280 publications

References 11 publications

Transfer Learning Through Greedy Subset Selection

Transfer Learning Through Greedy Subset Selection

Exploiting kernel-based feature weighting and instance clustering to transfer knowledge across domains

Cross-domain Text Classification with Multiple Domains and Disparate Label Sets

Contact Info

Product

Resources

About