Ensemble-Based Out-of-Distribution Detection
To design an efficient deep learning model that can be used in the real-world, it is important to detect out-of-distribution (OOD) data well. Various studies have been conducted to solve the OOD problem. The current state-of-the-art approach uses a confidence score based on the Mahalanobis distance...
Published in: | Electronics |
---|---|
Main Authors: | , , , , |
Format: | Article in Journal/Newspaper |
Language: | English |
Published: |
MDPI AG
2021
|
Subjects: | |
Online Access: | https://doi.org/10.3390/electronics10050567 https://doaj.org/article/2e94fb1c7e0d417fb89516e665c3b6fb |
id |
ftdoajarticles:oai:doaj.org/article:2e94fb1c7e0d417fb89516e665c3b6fb |
---|---|
record_format |
openpolar |
spelling |
ftdoajarticles:oai:doaj.org/article:2e94fb1c7e0d417fb89516e665c3b6fb 2024-01-07T09:42:55+01:00 Ensemble-Based Out-of-Distribution Detection Donghun Yang Kien Mai Ngoc Iksoo Shin Kyong-Ha Lee Myunggwon Hwang 2021-02-01T00:00:00Z https://doi.org/10.3390/electronics10050567 https://doaj.org/article/2e94fb1c7e0d417fb89516e665c3b6fb EN eng MDPI AG https://www.mdpi.com/2079-9292/10/5/567 https://doaj.org/toc/2079-9292 doi:10.3390/electronics10050567 2079-9292 https://doaj.org/article/2e94fb1c7e0d417fb89516e665c3b6fb Electronics, Vol 10, Iss 5, p 567 (2021) out-of-distribution detection confidence score distance metric learning Siamese network triplet network ensemble method Electronics TK7800-8360 article 2021 ftdoajarticles https://doi.org/10.3390/electronics10050567 2023-12-10T01:49:20Z To design an efficient deep learning model that can be used in the real-world, it is important to detect out-of-distribution (OOD) data well. Various studies have been conducted to solve the OOD problem. The current state-of-the-art approach uses a confidence score based on the Mahalanobis distance in a feature space. Although it outperformed the previous approaches, the results were sensitive to the quality of the trained model and the dataset complexity. Herein, we propose a novel OOD detection method that can train more efficient feature space for OOD detection. The proposed method uses an ensemble of the features trained using the softmax-based classifier and the network based on distance metric learning (DML). Through the complementary interaction of these two networks, the trained feature space has a more clumped distribution and can fit well on the Gaussian distribution by class. Therefore, OOD data can be efficiently detected by setting a threshold in the trained feature space. To evaluate the proposed method, we applied our method to various combinations of image datasets. The results show that the overall performance of the proposed approach is superior to those of other methods, including the state-of-the-art approach, on any combination of datasets. Article in Journal/Newspaper DML Directory of Open Access Journals: DOAJ Articles Electronics 10 5 567 |
institution |
Open Polar |
collection |
Directory of Open Access Journals: DOAJ Articles |
op_collection_id |
ftdoajarticles |
language |
English |
topic |
out-of-distribution detection confidence score distance metric learning Siamese network triplet network ensemble method Electronics TK7800-8360 |
spellingShingle |
out-of-distribution detection confidence score distance metric learning Siamese network triplet network ensemble method Electronics TK7800-8360 Donghun Yang Kien Mai Ngoc Iksoo Shin Kyong-Ha Lee Myunggwon Hwang Ensemble-Based Out-of-Distribution Detection |
topic_facet |
out-of-distribution detection confidence score distance metric learning Siamese network triplet network ensemble method Electronics TK7800-8360 |
description |
To design an efficient deep learning model that can be used in the real-world, it is important to detect out-of-distribution (OOD) data well. Various studies have been conducted to solve the OOD problem. The current state-of-the-art approach uses a confidence score based on the Mahalanobis distance in a feature space. Although it outperformed the previous approaches, the results were sensitive to the quality of the trained model and the dataset complexity. Herein, we propose a novel OOD detection method that can train more efficient feature space for OOD detection. The proposed method uses an ensemble of the features trained using the softmax-based classifier and the network based on distance metric learning (DML). Through the complementary interaction of these two networks, the trained feature space has a more clumped distribution and can fit well on the Gaussian distribution by class. Therefore, OOD data can be efficiently detected by setting a threshold in the trained feature space. To evaluate the proposed method, we applied our method to various combinations of image datasets. The results show that the overall performance of the proposed approach is superior to those of other methods, including the state-of-the-art approach, on any combination of datasets. |
format |
Article in Journal/Newspaper |
author |
Donghun Yang Kien Mai Ngoc Iksoo Shin Kyong-Ha Lee Myunggwon Hwang |
author_facet |
Donghun Yang Kien Mai Ngoc Iksoo Shin Kyong-Ha Lee Myunggwon Hwang |
author_sort |
Donghun Yang |
title |
Ensemble-Based Out-of-Distribution Detection |
title_short |
Ensemble-Based Out-of-Distribution Detection |
title_full |
Ensemble-Based Out-of-Distribution Detection |
title_fullStr |
Ensemble-Based Out-of-Distribution Detection |
title_full_unstemmed |
Ensemble-Based Out-of-Distribution Detection |
title_sort |
ensemble-based out-of-distribution detection |
publisher |
MDPI AG |
publishDate |
2021 |
url |
https://doi.org/10.3390/electronics10050567 https://doaj.org/article/2e94fb1c7e0d417fb89516e665c3b6fb |
genre |
DML |
genre_facet |
DML |
op_source |
Electronics, Vol 10, Iss 5, p 567 (2021) |
op_relation |
https://www.mdpi.com/2079-9292/10/5/567 https://doaj.org/toc/2079-9292 doi:10.3390/electronics10050567 2079-9292 https://doaj.org/article/2e94fb1c7e0d417fb89516e665c3b6fb |
op_doi |
https://doi.org/10.3390/electronics10050567 |
container_title |
Electronics |
container_volume |
10 |
container_issue |
5 |
container_start_page |
567 |
_version_ |
1787424180777517056 |