pith. sign in

arxiv: 2602.09295 · v2 · submitted 2026-02-10 · 💻 cs.LG · cs.SD

Positive-Unlabelled Active Learning to Curate a Dataset for Orca Resident Interpretation

Bret Nestor , Bohan Yao , Jasmine Moore , Jasper Kanes This is my paper

Pith reviewed 2026-05-16 03:39 UTC · model grok-4.3

classification 💻 cs.LG cs.SD
keywords Southern Resident Killer Whaleacoustic detectionpositive-unlabeled learningmarine mammal datasethydrophone archivestransformer classifierdataset curation
0
0 comments X

The pith

A positive-unlabelled active learning method extracts the largest public collection of Southern Resident Killer Whale acoustic data from over 30 years of archival hydrophone recordings.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper applies a weakly supervised strategy that begins with known positive examples of whale sounds and iteratively refines labels across vast unlabeled archives. This process yields hundreds of hours of cleaned marine mammal recordings, far exceeding prior public collections. Transformer models trained on the resulting data detect whale presence more accurately and with lower energy use than existing approaches on most test sets. The scale of the dataset opens direct applications in tracking endangered populations and surveying habitats. The authors release both the labels and pointers to the original audio under open licenses.

Core claim

By running positive-unlabelled active learning on weakly labeled archival hydrophone data, the work assembles 919 hours of Southern Resident Killer Whale recordings plus thousands of hours of other marine mammal sounds, trains a fleet of WHISPER-based presence classifiers that reach up to 0.77 AUROC, and produces multiclass species and ecotype models with 53.2 percent and 33.6 percent top-1 accuracy on held-out expert data.

What carries the argument

The positive-unlabelled active learning loop that starts from sparse weak positives and repeatedly selects uncertain instances for labeling within the archival audio streams.

Load-bearing premise

The iterative labeling process adds mostly true marine mammal calls rather than background noise or unrelated sounds.

What would settle it

Expert listening tests on a random sample of the newly labeled segments that measure the fraction incorrectly tagged as containing orca or other target species.

Figures

Figures reproduced from arXiv: 2602.09295 by Bohan Yao, Bret Nestor, Jasmine Moore, Jasper Kanes.

Figure 1
Figure 1. Figure 1: Hydrophone deployment sites where data are collected from. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Active learning with label noise. At each iteration [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Confusion matrix for the actual classes from DCLDE 2026 vs. the whisper-tiny species classifications. [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Confusion matrix for the actual ecotype classes from DCLDE 2026 vs. the whisper-tiny ecotype classifications. [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The criterion to determine which convergence regime is applicable. [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: AUROC across active learning iterations for a variety of settings [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: AUPRC across active learning iterations for a variety of settings [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
read the original abstract

This work presents the largest curation of Southern Resident Killer Whale (SRKW) acoustic data to date, also containing other marine mammals in their environment. We systematically search all available public archival hydrophone data within the SRKW habitat (over 30 years of audio data). The search consists of a weakly-supervised, positive-unlabelled, active learning strategy to identify all instances of marine mammals. The resulting transformer-based presence or absence classifiers outperform state-of-the-art classifiers on 3 of 4 expert-annotated datasets in terms of accuracy and energy efficiency. The fleet of WHISPER detection models range from 0.58 (0.48-0.67) AUROC with WHISPER-tiny to 0.77 (0.63-0.93) with WHISPER-large-v3. Our multiclass species classifier obtains a top-1 accuracy of 53.2\% (11 train classes, 4 test classes) and our ecotype classifier obtains a top-1 accuracy of 33.6\% (4 train classes, 5 test classes) on the DCLDE-2026 dataset. We yield 919 hours of SRKW data, 230 hours of Bigg's orca data, 1374 hours of orca data from unlabelled ecotypes, 1501 hours of humpback data, 88 hours of sea lion data, 246 hours of pacific white-sided dolphin data, and over 784 hours of unspecified marine mammal data. This SRKW dataset is larger than DCLDE-2026, Ocean Networks Canada, and OrcaSound combined. The curated species labels are available under CC-BY 4.0 license, and the corresponding audio data are available under the licenses of the original owners. The comprehensive nature of this dataset makes it suitable for unsupervised machine translation, habitat usage surveys, and conservation endeavours for this critically endangered ecotype.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper presents a positive-unlabeled active learning strategy applied to over 30 years of archival hydrophone data to curate the largest SRKW acoustic dataset to date (919 h SRKW, 1501 h humpback, etc.), along with other marine mammals. It claims the resulting transformer-based presence/absence classifiers (WHISPER fleet) outperform state-of-the-art on 3 of 4 expert-annotated datasets in accuracy and energy efficiency, with AUROCs ranging from 0.58 (0.48-0.67) for WHISPER-tiny to 0.77 (0.63-0.93) for WHISPER-large-v3; multiclass species accuracy is 53.2% (11 train/4 test classes) and ecotype accuracy is 33.6% (4 train/5 test classes) on DCLDE-2026. The curated labels are released under CC-BY 4.0.

Significance. If the label curation holds, the work makes a substantial contribution by releasing a large-scale, publicly available acoustic dataset exceeding prior collections (DCLDE-2026, Ocean Networks Canada, OrcaSound) and suitable for habitat surveys, conservation, and unsupervised translation tasks. The PU active learning approach on weakly labeled archival data is a practical innovation for marine acoustics, and the energy-efficiency claims for the WHISPER models could support edge deployment. Credit is due for the scale of curation and open release; however, significance is limited by the absence of independent label validation.

major comments (3)
  1. [Abstract and §4 (Results)] Abstract and §4 (Results): The claim that the classifiers outperform SOTA on 3/4 expert-annotated datasets in accuracy and AUROC rests on the assumption that the positive-unlabeled active learning pipeline yields sufficiently clean presence/absence labels from 30+ years of archival data. No quantification of label precision/recall or false-positive rate is provided against any held-out expert subset, which is load-bearing because contamination from background noise or non-target sounds would render the reported AUROCs (0.58–0.77) and accuracies (53.2%, 33.6%) uninterpretable.
  2. [Abstract] Abstract: The reported AUROC intervals (e.g., 0.58 (0.48-0.67) for WHISPER-tiny) are presented without details on computation method, whether they derive from multiple runs or cross-validation, data exclusion criteria, or how the active learning loop itself was evaluated on the expert-annotated sets.
  3. [§3 (Methods)] §3 (Methods): The positive-unlabeled active learning procedure is described at a high level but lacks an independent verification step (e.g., precision on a small expert-labeled hold-out from the archival data) that would confirm the training labels are clean enough to support the downstream performance claims.
minor comments (2)
  1. [Abstract] Abstract: The phrase 'fleet of WHISPER detection models' is used without specifying the exact model variants, fine-tuning protocol, or energy measurement methodology in the summary text.
  2. [§4 (Results)] The manuscript would benefit from a table summarizing the exact hours curated per species/ecotype alongside the corresponding expert-annotated test set sizes for direct comparison.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive feedback and for recognizing the scale of the curated dataset and its open release. We address each major comment below with proposed revisions to improve clarity and transparency around label quality and metric reporting.

read point-by-point responses

  1. Referee: [Abstract and §4 (Results)] Abstract and §4 (Results): The claim that the classifiers outperform SOTA on 3/4 expert-annotated datasets in accuracy and AUROC rests on the assumption that the positive-unlabeled active learning pipeline yields sufficiently clean presence/absence labels from 30+ years of archival data. No quantification of label precision/recall or false-positive rate is provided against any held-out expert subset, which is load-bearing because contamination from background noise or non-target sounds would render the reported AUROCs (0.58–0.77) and accuracies (53.2%, 33.6%) uninterpretable.

    Authors: We agree that direct quantification of label precision/recall on a held-out expert subset from the archival data would strengthen the claims and address potential concerns about contamination. The reported performance metrics are computed on independent expert-annotated test sets (DCLDE-2026 and others), providing evidence that the models generalize despite any training label noise. However, creating such a comprehensive held-out expert subset was not feasible due to annotation costs and scale. We will add a dedicated subsection in §3 or §4 describing the PU active learning validation steps used internally (e.g., iterative query selection metrics) and expand the limitations section to discuss the impact of possible label noise on the AUROCs and accuracies. revision: partial

  2. Referee: [Abstract] Abstract: The reported AUROC intervals (e.g., 0.58 (0.48-0.67) for WHISPER-tiny) are presented without details on computation method, whether they derive from multiple runs or cross-validation, data exclusion criteria, or how the active learning loop itself was evaluated on the expert-annotated sets.

    Authors: We will revise the abstract, §3, and §4 to explicitly state that the AUROC intervals are computed via bootstrap resampling (1000 iterations) over the test set samples, with no data exclusion beyond the predefined train/test splits. We will also clarify that the active learning loop was evaluated through the final models' performance on the held-out expert-annotated benchmarks rather than during the loop itself, and add details on any cross-validation used in hyperparameter tuning. revision: yes

  3. Referee: [§3 (Methods)] §3 (Methods): The positive-unlabeled active learning procedure is described at a high level but lacks an independent verification step (e.g., precision on a small expert-labeled hold-out from the archival data) that would confirm the training labels are clean enough to support the downstream performance claims.

    Authors: We agree the description is high-level and will expand §3 with additional details on the active learning iterations, including query selection criteria, stopping conditions, and any internal consistency checks performed. While a dedicated small expert-labeled hold-out from the archival data was not available (due to the 30+ year scale and expert time constraints), we will reference how the PU framework inherently reduces false positive risk and note the external benchmark results as supporting evidence of label utility. We will also add a brief discussion of this limitation. revision: partial

standing simulated objections not resolved
  • Direct numerical quantification of label precision/recall or false-positive rate on a held-out expert subset from the full archival dataset cannot be provided, as no such comprehensive expert annotations exist and creating them exceeds the scope and resources of this study.

Circularity Check

0 steps flagged

No circularity: empirical claims rest on external expert-annotated test sets

full rationale

The paper's central claims concern classifier performance (AUROC, accuracy) on four external expert-annotated datasets and the scale of the curated SRKW dataset. These quantities are obtained by direct evaluation against held-out expert labels that are independent of the positive-unlabeled active-learning pipeline used for curation. No equation, parameter fit, or self-citation is shown to define the reported performance numbers by construction; the derivation chain therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions of positive-unlabeled learning being applicable to noisy acoustic data and on the existence of sufficiently reliable expert-annotated test sets; no new free parameters, axioms, or invented entities are introduced beyond existing transformer architectures and PU-AL methods.

axioms (1)
  • domain assumption Positive-unlabeled learning assumptions hold for archival hydrophone recordings of marine mammals
    The method treats unlabeled data as containing both positive and negative examples without further justification in the abstract.

pith-pipeline@v0.9.0 · 5660 in / 1479 out tokens · 84956 ms · 2026-05-16T03:39:30.428806+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

73 extracted references · 73 canonical work pages · 1 internal anchor

  1. [1]

    Umang Aggarwal, Adrian Popescu, and Céline Hudelot. 2020. Active Learning for Imbalanced Datasets. In2020 IEEE Winter Conference on Applications of Computer Vision (W ACV). 1417–1426. doi:10.1109/WACV45572.2020.9093475

    work page doi:10.1109/wacv45572.2020.9093475 2020

  2. [2]

    Allen, Claire Garrigue, Emma L

    Inbal Arnon, Simon Kirby, Jenny A. Allen, Claire Garrigue, Emma L. Carroll, and Ellen C. Garland. 2025. Whale song shows language-like statistical structure.Science387, 6734 (2025), 649–

    work page 2025

  3. [3]

    arXiv:https://www.science.org/doi/pdf/10.1126/science.adq7055 doi:10.1126/science.adq7055

    work page doi:10.1126/science.adq7055

  4. [4]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. 2016. Layer normaliza- tion.arXiv preprint arXiv:1607.06450(2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  5. [5]

    Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Represen- tations. InAdvances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (Eds.), Vol. 33. Curran Associates, Inc., 12449–12460. https://proceedings.neurips.cc...

    work page 2020

  6. [6]

    Carretta, Shannon Rankin, Alexis Rudd, and Marc O

    Yvonne Barkley, Julie Nicola Oswald, James V. Carretta, Shannon Rankin, Alexis Rudd, and Marc O. Lammers. [2011. Comparison of real-time and post-cruise acoustic species identification of dolphin whistles using ROCCA (Real-time Odontocete Call Classification Algorithm). https://repository.library.noaa.gov/ view/noaa/3797 Technical Memorandum

    work page 2011

  7. [7]

    BARRETT-LENNARD, JOHN K.B

    LANCE G. BARRETT-LENNARD, JOHN K.B. FORD, and KATHY A. HEISE

    work page

  8. [8]

    Eyert, A comparative study on methods for convergence acceleration of iter- ative vector sequences , Journal of Computational Physics 124(2), 271 (1996), doi:10.1006/jcph.1996.0059

    The mixed blessing of echolocation: differences in sonar use by fish- eating and mammal-eating killer whales.Animal Behaviour51, 3 (1996), 553–565. doi:10.1006/anbe.1996.0059

    work page doi:10.1006/anbe.1996.0059 1996

  9. [9]

    Christian Bergler, Hendrik Schröter, Rachael Xi Cheng, Volker Barth, Michael Weber, Elmar Nöth, Heribert Hofer, and Andreas Maier. 2019. ORCA-SPOT: An Automatic Killer Whale Sound Detection Toolkit Using Deep Learning.Scientific Reports9, 1 (29 Jul 2019), 10997. doi:10.1038/s41598-019-47335-w

    work page doi:10.1038/s41598-019-47335-w 2019

  10. [10]

    Smeele, Stephen A

    Christian Bergler, Simeon Q. Smeele, Stephen A. Tyndel, Alexander Barnhill, Sara T. Ortiz, Ammie K. Kalan, Rachael Xi Cheng, Signe Brinkløv, Anna N. Osiecka, Jakob Tougaard, Freja Jakobsen, Magnus Wahlberg, Elmar Nöth, Andreas Maier, and Barbara C. Klump. 2022. ANIMAL-SPOT enables animal-independent signal detection and classification using deep learning....

    work page doi:10.1038/s41598-022-26429-y 2022

  11. [11]

    Michael Andrew Bigg. 1982. An assessment of killer whale (Orcinus orca) stocks off Vancouver Island, British Columbia.REP. INT. WHALING COMM32 (1982), 655–666. https://api.semanticscholar.org/CorpusID:222306729

    work page 1982

  12. [12]

    Alex Borowicz, Hieu Le, Grant Humphries, Georg Nehls, Caroline Höschle, Vladislav Kosarev, and Heather J. Lynch. 2019. Aerial-trained deep learning networks for surveying cetaceans from satellite imagery.PLOS ONE14, 10 (10 2019), 1–15. doi:10.1371/journal.pone.0212532

    work page doi:10.1371/journal.pone.0212532 2019

  13. [13]

    https://pubs.aip.org/asa/jasa/article- pdf/134/5/3486/13325241/3486_1_online.pdf

    M. Bradley Hanson, Candice K. Emmons, Eric J. Ward, Jeffrey A. Nystuen, and Marc O. Lammers. 2013. Assessing the coastal occur- rence of endangered killer whales using autonomous passive acous- tic recorders.The Journal of the Acoustical Society of America134, 5 (11 2013), 3486–3495. arXiv:"https://pubs.aip.org/asa/jasa/article- pdf/134/5/3486/13325241/34...

    work page doi:10.1121/1.4821206 2013

  14. [14]

    Olivier Coudray, Christine Keribin, Pascal Massart, and Patrick Pamphile. 2023. Risk Bounds for Positive-Unlabeled Learning Under the Selected At Random Assumption.Journal of Machine Learning Research24, 107 (2023), 1–31. http: //jmlr.org/papers/v24/22-067.html

    work page 2023

  15. [15]

    2024.mlco2/codecarbon: v2.4.1

    Benoit Courty, Victor Schmidt, Sasha Luccioni, Goyal-Kamal, MarionCoutarel, Boris Feld, Jérémy Lecourt, LiamConnell, Amine Saboni, Inimaz, supatomic, Mathilde Léval, Luis Blanche, Alexis Cruveiller, ouminasara, Franklin Zhao, Aditya Joshi, Alexis Bogroff, Hugues de Lavoreille, Niko Laskaris, Edoardo Abati, Douglas Blank, Ziyao Wang, Armin Catovic, Marc Al...

    work page doi:10.5281/zenodo.11171501 2024

  16. [16]

    Lusseau D, Bain DE, Williams R, and Smith JC. 2009. Vessel traffic disrupts the foraging behavior of southern resident killer whales Orcinus orca.Endangered Species Research6, 3 (2009), 211–221. doi:10.3354/esr00154

    work page doi:10.3354/esr00154 2009

  17. [17]

    Deecke, John K.B

    Volker B. Deecke, John K.B. Ford, and Peter J.B. Slater. 2005. The vocal behaviour of mammal-eating killer whales: communicating with costly calls.Animal Be- haviour69, 2 (2005), 395–405. doi:10.1016/j.anbehav.2004.04.014

    work page doi:10.1016/j.anbehav.2004.04.014 2005

  18. [18]

    1984.Call traditions and dialects of killer whales (Orcinus orca) in British Columbia

    John Kenneth Baker Ford. 1984.Call traditions and dialects of killer whales (Orcinus orca) in British Columbia. Ph. D. Dissertation. University of British Columbia. doi:10.14288/1.0096602

    work page doi:10.14288/1.0096602 1984

  19. [19]

    GBIF.Org User. 2024. Occurrence Download. doi:10.15468/DL.ERW6Z5

    work page doi:10.15468/dl.erw6z5 2024

  20. [20]

    Douglas Gillespie, Marjolaine Caillat, Jonathan Gordon, and Paul White. 2013. Automatic detection and classification of odontocete whistlesa).The Journal of the Acoustical Society of America134, 3 (09 2013), 2427–2437. arXiv:https://pubs.aip.org/asa/jasa/article- pdf/134/3/2427/15307906/2427_1_online.pdf doi:10.1121/1.4816555

    work page doi:10.1121/1.4816555 2013

  21. [21]

    Mellinger, Jonathan Gordon, David McLaren, Paul Redmond, Ronald McHugh, Philip Trinder, Xiao-Yan Deng, and Aaron Thode

    Douglas Gillespie, David K. Mellinger, Jonathan Gordon, David McLaren, Paul Redmond, Ronald McHugh, Philip Trinder, Xiao-Yan Deng, and Aaron Thode

    work page

  22. [22]

    doi:10.1121/1.4808713

    PAMGUARD: Semiautomated, open source software for real-time acoustic detection and localization of cetaceans.The Journal of the Acoustical Society of America125 (04 2009), 2547–2547. doi:10.1121/1.4808713

    work page doi:10.1121/1.4808713 2009

  23. [23]

    Bradley Hanson, Candice K

    M. Bradley Hanson, Candice K. Emmons, Michael J. Ford, Meredith Everett, Kim Parsons, Linda K. Park, Jennifer Hempelmann, Donald M. Van Doornik, Gregory S. Schorr, Jeffrey K. Jacobsen, Mark F. Sears, Maya S. Sears, John G. Sneva, Robin W. Baird, and Lynne Barre. 2021. Endangered predators and endangered prey: Seasonal diet of Southern Resident killer whal...

    work page doi:10.1371/journal.pone.0247031 2021

  24. [24]

    Leila Hatch, Jack Barkowski, Simone Baumann-Pickering, Carrie Wall Bell, Benjamin Best, Jennifer Brown, Annamaria DeAngelis, John Joseph, Ella Kim, Danielle Kitchen, Annebelle C. M. Kok, Anke Kügler, Anurag Kumar, Marc O. Lammers, Tetyana Margolina, Megan McKenna, Karlina Merkens, Lindsey Peavey Reeves, Timothy J. Rowell, Mandy Shoemaker, Jenni A. Stanley...

    work page doi:10.1007/978-3-031-50256-9_58 2024

  25. [25]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep Residual Learning for Image Recognition. In2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 770–778. doi:10.1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  26. [26]

    iNaturalist. [n. d.]. iNaturalist. https://www.inaturalist.org

    work page

  27. [27]

    NSF Ocean Observatories Initiative. 2023. Broadband Acoustic Receiver (Hydrophone) (CE02SHBP-LJ01D ) data from Oregon Shelf Cabled Ben- thic Experiment Package at Coastal Endurance from 2015-01-01 to 2018- 01-01. https://dataexplorer.oceanobservatories.org/#ooi/array/CE/subsite/ CE02SHBP/instrument_type/HYDBB/data

    work page 2023

  28. [28]

    IUCN. [n. d.]. IUCN Red List of Threatened Species. https://www.iucnredlist.org/

    work page

  29. [29]

    Joshi, Fatih Porikli, and Nikolaos Papanikolopoulos

    Ajay J. Joshi, Fatih Porikli, and Nikolaos Papanikolopoulos. 2009. Multi-class active learning for image classification. In2009 IEEE Conference on Computer Vision and Pattern Recognition. 2372–2379. doi:10.1109/CVPR.2009.5206627

    work page doi:10.1109/cvpr.2009.5206627 2009

  30. [30]

    2021 , issn =

    Stefan Kahl, Connor M. Wood, Maximilian Eibl, and Holger Klinck. 2021. BirdNET: A deep learning solution for avian diversity monitoring.Ecological Informatics 61 (2021), 101236. doi:10.1016/j.ecoinf.2021.101236

    work page doi:10.1016/j.ecoinf.2021.101236 2021

  31. [31]

    King and Vincent M

    Stephanie L. King and Vincent M. Janik. 2013. Bottlenose dol- phins can use learned vocal labels to address each other.Pro- ceedings of the National Academy of Sciences110, 32 (2013), 13216– 13221. arXiv:https://www.pnas.org/doi/pdf/10.1073/pnas.1304459110 doi:10.1073/pnas.1304459110

    work page doi:10.1073/pnas.1304459110 2013

  32. [32]

    Ryuichi Kiryo, Gang Niu, Marthinus C du Plessis, and Masashi Sugiyama. 2017. Positive-Unlabeled Learning with Non-Negative Risk Estimator. InAdvances in Neural Information Processing Systems, I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (Eds.), Vol. 30. Cur- ran Associates, Inc. https://proceedings.neurips.cc...

    work page 2017

  33. [33]

    Lacy, Rob Williams, Erin Ashe, Kenneth C

    Robert C. Lacy, Rob Williams, Erin Ashe, Kenneth C. Balcomb III, Lauren J. N. Brent, Christopher W. Clark, Darren P. Croft, Deborah A. Giles, Misty MacDuffee, and Paul C. Paquet. 2017. Evaluating anthropogenic threats to endangered killer whales to inform effective recovery plans.Scientific Reports7, 1 (26 Oct 2017), 14119. doi:10.1038/s41598-017-14471-0

    work page doi:10.1038/s41598-017-14471-0 2017

  34. [34]

    Guillaume Lample, Alexis Conneau, Marc’Aurelio Ranzato, Ludovic Denoyer, and Hervé Jégou. 2018. Word translation without parallel data. InInternational Con- ference on Learning Representations. https://openreview.net/forum?id=H196sainb

    work page 2018

  35. [35]

    Daniel Langenkämper, Erik Simon-Lledó, Brett Hosking, Daniel O. B. Jones, and Tim W. Nattkemper. 2019. On the impact of Citizen Science-derived data quality on deep learning based classification in marine images.PLOS ONE14, 6 (06 2019), 1–16. doi:10.1371/journal.pone.0218086

    work page doi:10.1371/journal.pone.0218086 2019

  36. [36]

    Lewis and William A

    David D. Lewis and William A. Gale. 1994. A sequential algorithm for training text classifiers. InProceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval(Dublin, Ireland)(SIGIR ’94). Springer-Verlag, Berlin, Heidelberg, 3–12

    work page 1994

  37. [37]

    Dongyuan Li, Zhen Wang, Yankai Chen, Renhe Jiang, Weiping Ding, and Manabu Okumura. 2025. A Survey on Deep Active Learning: Recent Advances and New Frontiers.IEEE Transactions on Neural Networks and Learning Systems36, 4 (2025), 5879–5899. doi:10.1109/TNNLS.2024.3396463

    work page doi:10.1109/tnnls.2024.3396463 2025

  38. [38]

    James, Maddie Cusimano, Masato Hagiwara, Sarah C

    Louis Mahon, Benjamin Hoffman, Logan S. James, Maddie Cusimano, Masato Hagiwara, Sarah C. Woolley, Felix Effenberger, Sara Keen, Jen-Yu Liu, and Olivier Pietquin. 2025. Robust detection of overlapping bioacoustic sound events. In The Thirty-Ninth Annual Conference on Neural Information Processing Systems workshop: AI for non-human animal communication. ht...

    work page 2025

  39. [39]

    MELLINGER, KATHLEEN M

    DAVID K. MELLINGER, KATHLEEN M. STAFFORD, SUE E. MOORE, ROBERT P. DZIAK, and HARU MATSUMOTO. 2007. An Overview of Fixed Passive Acoustic Observation Methods for Cetaceans.Oceanography20, 4 (2007), 36–45. http: //www.jstor.org/stable/24860138

    work page arXiv 2007

  40. [40]

    Morin, Morgan L

    Phillip A. Morin, Morgan L. McCarthy, Charissa W. Fung, John W. Dur- ban, Kim M. Parsons, William F. Perrin, Barbara L. Taylor, Thomas A. Jef- ferson, and Frederick I. Archer. 2024. Revised taxonomy of eastern North Positive-Unlabelled Active Learning to Curate a Dataset for Orca Resident Interpretation Under review for KDD ’26, August 09–13, 2026, Jeju, ...

    work page doi:10.1098/rsos.231368 2024

  41. [41]

    Nagarajan Natarajan, Inderjit S Dhillon, Pradeep K Ravikumar, and Ambuj Tewari

    work page

  42. [42]

    InAdvances in Neural Information Processing Systems, C.J

    Learning with Noisy Labels. InAdvances in Neural Information Processing Systems, C.J. Burges, L. Bottou, M. Welling, Z. Ghahramani, and K.Q. Weinberger (Eds.), Vol. 26. Curran Associates, Inc. https://proceedings.neurips.cc/paper_ files/paper/2013/file/3871bd64012152bfb53fdf04b401193f-Paper.pdf

    work page 2013

  43. [43]

    Orca Network. [n. d.]. Recent Whale Sightings. https://www.orcanetwork.org/ recent-sightings. Accessed: 2023-10-14

    work page 2023

  44. [44]

    Kun-Peng Ning, Xun Zhao, Yu Li, and Sheng-Jun Huang. 2022. Active Learning for Open-Set Annotation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 41–49

    work page 2022

  45. [45]

    Ocean Networks Canada Society. 2020. Folger Deep Hydrophone Deployed 2020-03-08. doi:10.34943/8aa1dc24-2e86-45ad-a46b-6ed1c9a3e0df

    work page doi:10.34943/8aa1dc24-2e86-45ad-a46b-6ed1c9a3e0df 2020

  46. [46]

    Ocean Networks Canada Society. 2023. Strait of Georgia Central Hydrophone Deployed 2023-03-17. doi:10.34943/b169fbab-3287-44f0-b5d6-a27cb2d6b1a0

    work page doi:10.34943/b169fbab-3287-44f0-b5d6-a27cb2d6b1a0 2023

  47. [47]

    National Oceanic and Department of Commerce Atmospheric Administration National Marine Fisheries Service. [n. d.]. Enumeration of endangered marine and anadromous species. 50 CFR 224.101. https://www.ecfr.gov/current/title- 50/part-224/section-224.101

    work page

  48. [48]

    Government of Canada. 2023. Killer Whale (Orcinus orca), Northeast Pacific southern resident population. https://species-registry.canada.ca/index-en.html# /species/699-5

    work page 2023

  49. [49]

    K. B. Oke, C. J. Cunningham, P. A. H. Westley, M. L. Baskett, S. M. Carlson, J. Clark, A. P. Hendry, V. A. Karatayev, N. W. Kendall, J. Kibele, H. K. Kindsvater, K. M. Kobayashi, B. Lewis, S. Munch, J. D. Reynolds, G. K. Vick, and E. P. Palkovacs. 2020. Recent declines in salmon body size impact ecosystems and fisheries.Nature Communications11, 1 (19 Aug ...

    work page doi:10.1038/s41467-020-17726-z 2020

  50. [50]

    OrcaSound. [n. d.]. Projects. https://www.orcasound.net/portfolio/. Accessed: 2023-10-14

    work page 2023

  51. [51]

    K. J. Palmer, Emma Cummings, Michael G. Dowd, Kait Frasier, Fabio Frazao, Alex Harris, April Houweling, Jasper Kanes, Oliver S. Kirsebom, Holger Klinck, Holly LeBlond, Lauren Laturnus, Craig Matkin, Olivia Murphy, Hannah Myers, Dan Olsen, Caitlin O’Neill, Bruno Padovese, James Pilkington, Lucy Quayle, Amalis Riera Vuibert, Krista Trounce, Svein Vagle, Sco...

    work page doi:10.1038/s41597-025- 2025

  52. [52]

    Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. 2015. Librispeech: An ASR corpus based on public domain audio books. In2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 5206–5210. doi:10.1109/ICASSP.2015.7178964

    work page doi:10.1109/icassp.2015.7178964 2015

  53. [53]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cour- napeau, M. Brucher, M. Perrot, and E. Duchesnay. 2011. Scikit-learn: Machine Learning in Python.Journal of Machine Learning Research12 (2011), 2825–2830

    work page 2011

  54. [54]

    Pilkington, Eva H

    James F. Pilkington, Eva H. Stredulinsky, Katherine Gavrilchuk, Sheila J. Thornton, John K. B. Ford, and Thomas Doniol-Valcroze. 2023. Patterns of winter occurrence of three sympatric killer whale populations off eastern Vancouver Island, Canada, based on passive acoustic monitoring.Frontiers in Marine ScienceVolume 10 - 2023 (2023). doi:10.3389/fmars.202...

    work page doi:10.3389/fmars.2023.1204908 2023

  55. [55]

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine Mcleavey, and Ilya Sutskever. 2023. Robust Speech Recognition via Large-Scale Weak Super- vision. InProceedings of the 40th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 202), Andreas Krause, Emma Brun- skill, Kyunghyun Cho, Barbara Engelhardt, Siv...

    work page 2023

  56. [57]

    Rand, Jason D

    Zoe R. Rand, Jason D. Wood, and Julie N. Oswald. 2022. Effects of duty cy- cles on passive acoustic monitoring of southern resident killer whale (Orci- nus orca) occurrence and behavior.The Journal of the Acoustical Society of America151, 3 (03 2022), 1651–1660. arXiv:https://pubs.aip.org/asa/jasa/article- pdf/151/3/1651/16523645/1651_1_online.pdf doi:10....

    work page doi:10.1121/10.0009752 2022

  57. [58]

    Leger, Dawn P

    Stephen Raverty, Judy St. Leger, Dawn P. Noren, Kathy Burek Huntington, David S. Rotstein, Frances M. D. Gulland, John K. B. Ford, M. Bradley Hanson, Dyanna M. Lambourn, Jessie Huggins, Martha A. Delaney, Lisa Spaven, Teri Rowles, Lynne Barre, Paul Cottrell, Graeme Ellis, Tracey Goldstein, Karen Terio, Debbie Duffield, Jim Rice, and Joseph K. Gaydos. 2020...

    work page 2020

  58. [59]

    https://pubs.aip.org/asa/jasa/article- pdf/134/3/2393/15307055/2393_1_online.pdf

    Amalis Riera, John K. Ford, and N. Ross Chapman. 2013. Effects of dif- ferent analysis techniques and recording duty cycles on passive acoustic monitoring of killer whales.The Journal of the Acoustical Society of Amer- ica134, 3 (09 2013), 2393–2404. arXiv:"https://pubs.aip.org/asa/jasa/article- pdf/134/3/2393/15307055/2393_1_online.pdf" doi:10.1121/1.4816552

    work page doi:10.1121/1.4816552 2013

  59. [60]

    Riera, J

    A. Riera, J. F. Pilkington, J. K. B. Ford, E. H. Stredulinsky, and N. R. Chapman

    work page

  60. [61]

    doi:10.3354/esr00966

    Passive acoustic monitoring off Vancouver Island reveals extensive use by at-risk Resident killer whale (Orcinus orca) populations.Endangered Species Research39 (2019), 221–234. doi:10.3354/esr00966

    work page doi:10.3354/esr00966 2019

  61. [62]

    Gruber, Daniela Rus, An- tonio Torralba, and Jacob Andreas

    Pratyusha Sharma, Shane Gero, Roger Payne, David F. Gruber, Daniela Rus, An- tonio Torralba, and Jacob Andreas. 2024. Contextual and combinatorial structure in sperm whale vocalisations.Nature Communications15, 1 (07 May 2024), 3617. doi:10.1038/s41467-024-47221-8

    work page doi:10.1038/s41467-024-47221-8 2024

  62. [63]

    Monika W Shields. 2023. 2018–2022 Southern Resident killer whale presence in the Salish Sea: continued shifts in habitat usage.PeerJ11 (2023), e15635. doi:10.7717/peerj.15635

    work page doi:10.7717/peerj.15635 2023

  63. [64]

    Marie Souhaut and Monika W Shields. 2021. Stereotyped whistles in southern resident killer whales.PeerJ9, e12085 (Aug. 2021), e12085

    work page 2021

  64. [65]

    Stewart, Jane Cogan, John W

    Joshua D. Stewart, Jane Cogan, John W. Durban, Holly Fearnbach, David K. Ellifrit, Mark Malleson, Melisa Pinnow, and Kenneth C. Balcomb. 2023. Traditional summer habitat use by Southern Resident killer whales in the Salish Sea is linked to Fraser River Chinook salmon returns.Marine Mammal Science39, 3 (2023), 858–

    work page 2023

  65. [66]

    D., Cogan, J., Durban, J

    arXiv:https://onlinelibrary.wiley.com/doi/pdf/10.1111/mms.13012 doi:10. 1111/mms.13012

    work page doi:10.1111/mms.13012

  66. [67]

    Tennessen, Marla M

    Jennifer B. Tennessen, Marla M. Holt, Brianna M. Wright, M. Bradley Han- son, Candice K. Emmons, Deborah A. Giles, Jeffrey T. Hogan, Sheila J. Thornton, and Volker B. Deecke. 2024. Males miss and females forgo: Auditory masking from vessel noise impairs foraging efficiency and suc- cess in killer whales.Global Change Biology30, 9 (2024), e17490. arXiv:htt...

    work page doi:10.1111/gcb.17490 2024

  67. [68]

    George Tzanetakis, Mathieu Lagrange, Paul Spong, and Helena Symonds. 2007. Orchive: Digitizing and Analyzing Orca Vocalizations. InLarge Scale Semantic Access to Content (Text, Image, Video, and Sound)(Pittsburgh, Pennsylvania)(RIAO ’07). LE CENTRE DE HAUTES ETUDES INTERNATIONALES D’INFORMATIQUE DOCUMENTAIRE, Paris, FRA, 56–63

    work page 2007

  68. [69]

    2021.Vessel presence and acoustic environment within Southern Resident killer whale (Orcinus orca) critical habitat in the Salish Sea and Swiftsure Bank area

    Svein Vagle, Rianna Burnham, Pramod Thupaki, Christine Konrad, Scott Toews, and Sheila J Thornton. 2021.Vessel presence and acoustic environment within Southern Resident killer whale (Orcinus orca) critical habitat in the Salish Sea and Swiftsure Bank area. Canadian Science Advisory Secretariat (CSAS)

    work page 2021

  69. [70]

    Vigness-Raposa, Gail Scowcroft, James H

    Kathleen J. Vigness-Raposa, Gail Scowcroft, James H. Miller, and Darlene Ketten

    work page

  70. [71]

    InThe Effects of Noise on Aquatic Life, Arthur N

    Discovery of Sound in the Sea: An Online Resource. InThe Effects of Noise on Aquatic Life, Arthur N. Popper and Anthony Hawkins (Eds.). Springer New York, New York, NY, 135–138

    work page

  71. [72]

    Rob Williams, David Lusseau, and Philip S. Hammond. 2006. Estimating relative energetic costs of human disturbance to killer whales (Orcinus orca).Biological Conservation133, 3 (2006), 301–311. doi:10.1016/j.biocon.2006.06.010

    work page doi:10.1016/j.biocon.2006.06.010 2006

  72. [73]

    Donggeun Yoo and In So Kweon. 2019. Learning Loss for Active Learning. In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 93–102. doi:10.1109/CVPR.2019.00018

    work page doi:10.1109/cvpr.2019.00018 2019

  73. [74]

    Chulhee Yun, Srinadh Bhojanapalli, Ankit Singh Rawat, Sashank Reddi, and Sanjiv Kumar. 2020. Are Transformers universal approximators of sequence- to-sequence functions?. InInternational Conference on Learning Representations. https://openreview.net/forum?id=ByxRM0Ntvr Under review for KDD ’26, August 09–13, 2026, Jeju, Korea Nestor et al. A Appendix Rece...

    work page 2020