Using context to train time-domain echolocation click detectorsa)
This work demonstrates the effectiveness of using humans in the loop processes for constructing large training sets for machine learning tasks. A corpus of over 57 000 toothed whale echolocation clicks was developed by using a permissive energy-based echolocation detector followed by a machine-assis...
Main Authors: | , , , , , , |
---|---|
Format: | Article in Journal/Newspaper |
Language: | unknown |
Published: |
eScholarship, University of California
2021
|
Subjects: | |
Online Access: | https://escholarship.org/uc/item/2j76x8bp |
id |
ftcdlib:oai:escholarship.org:ark:/13030/qt2j76x8bp |
---|---|
record_format |
openpolar |
spelling |
ftcdlib:oai:escholarship.org:ark:/13030/qt2j76x8bp 2023-11-05T03:45:23+01:00 Using context to train time-domain echolocation click detectorsa) Roch, Marie A Lindeneau, Scott Aurora, Gurisht Singh Frasier, Kaitlin E Hildebrand, John A Glotin, Hervé Baumann-Pickering, Simone 3301 - 3310 2021-05-01 https://escholarship.org/uc/item/2j76x8bp unknown eScholarship, University of California qt2j76x8bp https://escholarship.org/uc/item/2j76x8bp public The Journal of the Acoustical Society of America, vol 149, iss 5 Information and Computing Sciences Machine Learning Animals Cetacea Echolocation Neural Networks Computer Vocalization Animal Acoustics article 2021 ftcdlib 2023-10-09T18:04:56Z This work demonstrates the effectiveness of using humans in the loop processes for constructing large training sets for machine learning tasks. A corpus of over 57 000 toothed whale echolocation clicks was developed by using a permissive energy-based echolocation detector followed by a machine-assisted quality control process that exploits contextual cues. Subsets of these data were used to train feed forward neural networks that detected over 850 000 echolocation clicks that were validated using the same quality control process. It is shown that this network architecture performs well in a variety of contexts and is evaluated against a withheld data set that was collected nearly five years apart from the development data at a location over 600 km distant. The system was capable of finding echolocation bouts that were missed by human analysts, and the patterns of error in the classifier consist primarily of anthropogenic sources that were not included as counter-training examples. In the absence of such events, typical false positive rates are under ten events per hour even at low thresholds. Article in Journal/Newspaper toothed whale University of California: eScholarship |
institution |
Open Polar |
collection |
University of California: eScholarship |
op_collection_id |
ftcdlib |
language |
unknown |
topic |
Information and Computing Sciences Machine Learning Animals Cetacea Echolocation Neural Networks Computer Vocalization Animal Acoustics |
spellingShingle |
Information and Computing Sciences Machine Learning Animals Cetacea Echolocation Neural Networks Computer Vocalization Animal Acoustics Roch, Marie A Lindeneau, Scott Aurora, Gurisht Singh Frasier, Kaitlin E Hildebrand, John A Glotin, Hervé Baumann-Pickering, Simone Using context to train time-domain echolocation click detectorsa) |
topic_facet |
Information and Computing Sciences Machine Learning Animals Cetacea Echolocation Neural Networks Computer Vocalization Animal Acoustics |
description |
This work demonstrates the effectiveness of using humans in the loop processes for constructing large training sets for machine learning tasks. A corpus of over 57 000 toothed whale echolocation clicks was developed by using a permissive energy-based echolocation detector followed by a machine-assisted quality control process that exploits contextual cues. Subsets of these data were used to train feed forward neural networks that detected over 850 000 echolocation clicks that were validated using the same quality control process. It is shown that this network architecture performs well in a variety of contexts and is evaluated against a withheld data set that was collected nearly five years apart from the development data at a location over 600 km distant. The system was capable of finding echolocation bouts that were missed by human analysts, and the patterns of error in the classifier consist primarily of anthropogenic sources that were not included as counter-training examples. In the absence of such events, typical false positive rates are under ten events per hour even at low thresholds. |
format |
Article in Journal/Newspaper |
author |
Roch, Marie A Lindeneau, Scott Aurora, Gurisht Singh Frasier, Kaitlin E Hildebrand, John A Glotin, Hervé Baumann-Pickering, Simone |
author_facet |
Roch, Marie A Lindeneau, Scott Aurora, Gurisht Singh Frasier, Kaitlin E Hildebrand, John A Glotin, Hervé Baumann-Pickering, Simone |
author_sort |
Roch, Marie A |
title |
Using context to train time-domain echolocation click detectorsa) |
title_short |
Using context to train time-domain echolocation click detectorsa) |
title_full |
Using context to train time-domain echolocation click detectorsa) |
title_fullStr |
Using context to train time-domain echolocation click detectorsa) |
title_full_unstemmed |
Using context to train time-domain echolocation click detectorsa) |
title_sort |
using context to train time-domain echolocation click detectorsa) |
publisher |
eScholarship, University of California |
publishDate |
2021 |
url |
https://escholarship.org/uc/item/2j76x8bp |
op_coverage |
3301 - 3310 |
genre |
toothed whale |
genre_facet |
toothed whale |
op_source |
The Journal of the Acoustical Society of America, vol 149, iss 5 |
op_relation |
qt2j76x8bp https://escholarship.org/uc/item/2j76x8bp |
op_rights |
public |
_version_ |
1781707441818304512 |