Distance Metric Learning Using Dropout: A Structured Regularization Approach
Distance metric learning (DML) aims to learn a distance metric better than Euclidean distance. It has been successfully applied to various tasks, e.g., classification, clustering and information retrieval. Many DML algorithms suffer from the over-fitting problem because of a large number of paramete...
Published in: | Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining |
---|---|
Main Authors: | , , , , |
Format: | Text |
Language: | unknown |
Published: |
UW Tacoma Digital Commons
2014
|
Subjects: | |
Online Access: | https://digitalcommons.tacoma.uw.edu/tech_pub/205 https://doi.org/10.1145/2623330.2623678 |
id |
ftuniwashingtaco:oai:digitalcommons.tacoma.uw.edu:tech_pub-1204 |
---|---|
record_format |
openpolar |
spelling |
ftuniwashingtaco:oai:digitalcommons.tacoma.uw.edu:tech_pub-1204 2023-09-05T13:19:04+02:00 Distance Metric Learning Using Dropout: A Structured Regularization Approach Qian, Qi Hu, Juhua Jin, Rong Pei, Jian Zhu, Shenghuo 2014-01-01T08:00:00Z https://digitalcommons.tacoma.uw.edu/tech_pub/205 https://doi.org/10.1145/2623330.2623678 unknown UW Tacoma Digital Commons https://digitalcommons.tacoma.uw.edu/tech_pub/205 doi:10.1145/2623330.2623678 School of Engineering and Technology Publications distance metric learning dropout text 2014 ftuniwashingtaco https://doi.org/10.1145/2623330.2623678 2023-08-21T14:16:50Z Distance metric learning (DML) aims to learn a distance metric better than Euclidean distance. It has been successfully applied to various tasks, e.g., classification, clustering and information retrieval. Many DML algorithms suffer from the over-fitting problem because of a large number of parameters to be determined in DML. In this paper, we exploit the dropout technique, which has been successfully applied in deep learning to alleviate the over-fitting problem, for DML. Different from the previous studies that only apply dropout to training data, we apply dropout to both the learned metrics and the training data. We illustrate that application of dropout to DML is essentially equivalent to matrix norm based regularization. Compared with the standard regularization scheme in DML, dropout is advantageous in simulating the structured regularizers which have shown consistently better performance than non structured regularizers. We verify, both empirically and theoretically, that dropout is effective in regulating the learned metric to avoid the over-fitting problem. Last, we examine the idea of wrapping the dropout technique in the state-of-art DML methods and observe that the dropout technique can significantly improve the performance of the original DML methods. Text DML University of Washington: UW Tacoma Digital Commons Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining 323 332 |
institution |
Open Polar |
collection |
University of Washington: UW Tacoma Digital Commons |
op_collection_id |
ftuniwashingtaco |
language |
unknown |
topic |
distance metric learning dropout |
spellingShingle |
distance metric learning dropout Qian, Qi Hu, Juhua Jin, Rong Pei, Jian Zhu, Shenghuo Distance Metric Learning Using Dropout: A Structured Regularization Approach |
topic_facet |
distance metric learning dropout |
description |
Distance metric learning (DML) aims to learn a distance metric better than Euclidean distance. It has been successfully applied to various tasks, e.g., classification, clustering and information retrieval. Many DML algorithms suffer from the over-fitting problem because of a large number of parameters to be determined in DML. In this paper, we exploit the dropout technique, which has been successfully applied in deep learning to alleviate the over-fitting problem, for DML. Different from the previous studies that only apply dropout to training data, we apply dropout to both the learned metrics and the training data. We illustrate that application of dropout to DML is essentially equivalent to matrix norm based regularization. Compared with the standard regularization scheme in DML, dropout is advantageous in simulating the structured regularizers which have shown consistently better performance than non structured regularizers. We verify, both empirically and theoretically, that dropout is effective in regulating the learned metric to avoid the over-fitting problem. Last, we examine the idea of wrapping the dropout technique in the state-of-art DML methods and observe that the dropout technique can significantly improve the performance of the original DML methods. |
format |
Text |
author |
Qian, Qi Hu, Juhua Jin, Rong Pei, Jian Zhu, Shenghuo |
author_facet |
Qian, Qi Hu, Juhua Jin, Rong Pei, Jian Zhu, Shenghuo |
author_sort |
Qian, Qi |
title |
Distance Metric Learning Using Dropout: A Structured Regularization Approach |
title_short |
Distance Metric Learning Using Dropout: A Structured Regularization Approach |
title_full |
Distance Metric Learning Using Dropout: A Structured Regularization Approach |
title_fullStr |
Distance Metric Learning Using Dropout: A Structured Regularization Approach |
title_full_unstemmed |
Distance Metric Learning Using Dropout: A Structured Regularization Approach |
title_sort |
distance metric learning using dropout: a structured regularization approach |
publisher |
UW Tacoma Digital Commons |
publishDate |
2014 |
url |
https://digitalcommons.tacoma.uw.edu/tech_pub/205 https://doi.org/10.1145/2623330.2623678 |
genre |
DML |
genre_facet |
DML |
op_source |
School of Engineering and Technology Publications |
op_relation |
https://digitalcommons.tacoma.uw.edu/tech_pub/205 doi:10.1145/2623330.2623678 |
op_doi |
https://doi.org/10.1145/2623330.2623678 |
container_title |
Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining |
container_start_page |
323 |
op_container_end_page |
332 |
_version_ |
1776199878878167040 |