Details
| Original language | English |
|---|---|
| Pages (from-to) | 845-852 |
| Number of pages | 8 |
| Journal | International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives |
| Volume | 42 |
| Issue number | 2 |
| Publication status | Published - 30 May 2018 |
| Event | 2018 ISPRS TC II Mid-term Symposium "Towards Photogrammetry 2020" - Riva del Garda, Italy Duration: 4 Jun 2018 → 7 Jun 2018 |
Abstract
In this paper we deal with the problem of measuring the similarity between training and tests datasets in the context of transfer learning (TL) for image classification. TL tries to transfer knowledge from a source domain, where labelled training samples are abundant but the data may follow a different distribution, to a target domain, where labelled training samples are scarce or even unavailable, assuming that the domains are related. Thus, the requirements w.r.t. the availability of labelled training samples in the target domain are reduced. In particular, if no labelled target data are available, it is inherently difficult to find a robust measure of relatedness between the source and target domains. This is of crucial importance for the performance of TL, because the knowledge transfer between unrelated data may lead to negative transfer, i.e. to a decrease of classification performance after transfer. We address the problem of measuring the relatedness between source and target datasets and investigate three different strategies to predict and, consequently, to avoid negative transfer in this paper. The first strategy is based on circular validation. The second strategy relies on the Maximum Mean Discrepancy (MMD) similarity metric, whereas the third one is an extension of MMD which incorporates the knowledge about the class labels in the source domain. Our method is evaluated using two different benchmark datasets. The experiments highlight the strengths and weaknesses of the investigated methods. We also show that it is possible to reduce the amount of negative transfer using these strategies for a TL method and to generate a consistent performance improvement over the whole dataset.
Keywords
- Domain adaptation, Negative transfer, Remote sensing, Transfer learning
ASJC Scopus subject areas
- Computer Science(all)
- Information Systems
- Social Sciences(all)
- Geography, Planning and Development
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives, Vol. 42, No. 2, 30.05.2018, p. 845-852.
Research output: Contribution to journal › Conference article › Research › peer review
}
TY - JOUR
T1 - A comparison of two strategies for avoiding negative transfer in domain adaptation based on logistic regression
AU - Paul, A.
AU - Vogt, K.
AU - Rottensteiner, F.
AU - Ostermann, J.
AU - Heipke, C.
N1 - Funding information: This work was supported by the German Science Foundation (DFG) under grant HE 1822/30-1. The Vaihingen and Potsdam data were provided by the German Society for Photogrammetry, Remote Sensing and Geoinformation (DGPF) (Cramer, 2010): http://www.ifp.uni-stuttgart.de/dgpf/DKEP-Allg.html.
PY - 2018/5/30
Y1 - 2018/5/30
N2 - In this paper we deal with the problem of measuring the similarity between training and tests datasets in the context of transfer learning (TL) for image classification. TL tries to transfer knowledge from a source domain, where labelled training samples are abundant but the data may follow a different distribution, to a target domain, where labelled training samples are scarce or even unavailable, assuming that the domains are related. Thus, the requirements w.r.t. the availability of labelled training samples in the target domain are reduced. In particular, if no labelled target data are available, it is inherently difficult to find a robust measure of relatedness between the source and target domains. This is of crucial importance for the performance of TL, because the knowledge transfer between unrelated data may lead to negative transfer, i.e. to a decrease of classification performance after transfer. We address the problem of measuring the relatedness between source and target datasets and investigate three different strategies to predict and, consequently, to avoid negative transfer in this paper. The first strategy is based on circular validation. The second strategy relies on the Maximum Mean Discrepancy (MMD) similarity metric, whereas the third one is an extension of MMD which incorporates the knowledge about the class labels in the source domain. Our method is evaluated using two different benchmark datasets. The experiments highlight the strengths and weaknesses of the investigated methods. We also show that it is possible to reduce the amount of negative transfer using these strategies for a TL method and to generate a consistent performance improvement over the whole dataset.
AB - In this paper we deal with the problem of measuring the similarity between training and tests datasets in the context of transfer learning (TL) for image classification. TL tries to transfer knowledge from a source domain, where labelled training samples are abundant but the data may follow a different distribution, to a target domain, where labelled training samples are scarce or even unavailable, assuming that the domains are related. Thus, the requirements w.r.t. the availability of labelled training samples in the target domain are reduced. In particular, if no labelled target data are available, it is inherently difficult to find a robust measure of relatedness between the source and target domains. This is of crucial importance for the performance of TL, because the knowledge transfer between unrelated data may lead to negative transfer, i.e. to a decrease of classification performance after transfer. We address the problem of measuring the relatedness between source and target datasets and investigate three different strategies to predict and, consequently, to avoid negative transfer in this paper. The first strategy is based on circular validation. The second strategy relies on the Maximum Mean Discrepancy (MMD) similarity metric, whereas the third one is an extension of MMD which incorporates the knowledge about the class labels in the source domain. Our method is evaluated using two different benchmark datasets. The experiments highlight the strengths and weaknesses of the investigated methods. We also show that it is possible to reduce the amount of negative transfer using these strategies for a TL method and to generate a consistent performance improvement over the whole dataset.
KW - Domain adaptation
KW - Negative transfer
KW - Remote sensing
KW - Transfer learning
UR - http://www.scopus.com/inward/record.url?scp=85048360278&partnerID=8YFLogxK
U2 - 10.5194/isprs-archives-XLII-2-845-2018
DO - 10.5194/isprs-archives-XLII-2-845-2018
M3 - Conference article
AN - SCOPUS:85048360278
VL - 42
SP - 845
EP - 852
JO - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives
JF - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives
SN - 1682-1750
IS - 2
T2 - 2018 ISPRS TC II Mid-term Symposium "Towards Photogrammetry 2020"
Y2 - 4 June 2018 through 7 June 2018
ER -