{"paper":{"title":"Masked Autoencoders with Limited Data: Does It Work? A Fine-Grained Bioacoustics Case Study","license":"http://creativecommons.org/licenses/by/4.0/","headline":"For fine-grained bioacoustic classification with limited labels, pretraining on large general audio datasets beats additional domain-specific masked autoencoder training.","cross_cats":["cs.CV","cs.LG"],"primary_cat":"cs.SD","authors_text":"Grant Van Horn, Mustafa Chasmai, Subhransu Maji, Wuao Liu","submitted_at":"2026-05-13T18:45:08Z","abstract_excerpt":"Bioacoustic recognition requires fine-grained acoustic understanding to distinguish similar-sounding species. However, many large-scale data repositories such as iNaturalist are weakly annotated, often with only a single positive species label per recording, making supervised learning particularly challenging. Inspired by advances in computer vision, recent approaches have shifted toward self-supervised learning to capture the underlying structure of audio without relying on exhaustive annotations. In particular, masked autoencoders (MAE) have shown strong transferability on massive audio corp"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"In moderate-sized fine-grained bioacoustic settings, pretraining scale dominates objective design.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That performance differences observed on iNatSounds are driven primarily by pretraining data scale rather than uncontrolled factors such as exact model capacity, optimizer choices, or dataset-specific biases in the weakly labeled recordings.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"In moderate-sized fine-grained bioacoustics, pretraining scale of masked autoencoders on diverse general audio dominates over domain-specific objectives or data curation for transfer performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"For fine-grained bioacoustic classification with limited labels, pretraining on large general audio datasets beats additional domain-specific masked autoencoder training.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2225fd34e9f3c00c8a77410a9d91271017cb2305d8c5c51379bb6f33a3a79cac"},"source":{"id":"2605.14031","kind":"arxiv","version":1},"verdict":{"id":"46b520e4-8778-46cb-994b-52398184b698","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:41:23.006797Z","strongest_claim":"In moderate-sized fine-grained bioacoustic settings, pretraining scale dominates objective design.","one_line_summary":"In moderate-sized fine-grained bioacoustics, pretraining scale of masked autoencoders on diverse general audio dominates over domain-specific objectives or data curation for transfer performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That performance differences observed on iNatSounds are driven primarily by pretraining data scale rather than uncontrolled factors such as exact model capacity, optimizer choices, or dataset-specific biases in the weakly labeled recordings.","pith_extraction_headline":"For fine-grained bioacoustic classification with limited labels, pretraining on large general audio datasets beats additional domain-specific masked autoencoder training."},"references":{"count":58,"sample":[{"doi":"","year":2022,"title":"Mae-ast: Masked autoencoding audio spectrogram transformer","work_id":"0d9a5b49-c744-4674-b38c-2e06181f73dd","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Entropy-based analysis of influential factors for underwater acoustic target recognition in passive sonar data.Ocean Engineering, 342: 122908, 2025","work_id":"20e035c4-b9b4-4aa4-8370-2fde94c6de85","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"wav2vec 2.0: A framework for self-supervised learning of speech representations.Advances in neural infor- mation processing systems, 33:12449–12460, 2020","work_id":"163aa9cd-370e-46d2-8cfb-63f1a4398ac9","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2010,"title":"Global biodiversity: indicators of recent declines","work_id":"a16884ff-e7e5-4560-b4d3-d4b9371a1c59","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"The inaturalist sounds dataset.Advances in Neural Information Processing Systems, 37:132524–132544,","work_id":"326b5f90-3748-45bd-bd24-ac7d5aecf34e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":58,"snapshot_sha256":"3134df022861d7102da2c65801d8f5b53e6b4e338c2412878bdf767d920672da","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4b8110fc7eb73fa247e641717bf2b9d5bc25cb9443e54c9d15f662efc116a0c1"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}