{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TNERJ53SRN5J7ABKFKDW4W7VLU","short_pith_number":"pith:TNERJ53S","schema_version":"1.0","canonical_sha256":"9b4914f7728b7a9f802a2a876e5bf55d24c382f9ae46aea0b3e26090c475a8c7","source":{"kind":"arxiv","id":"2509.24901","version":4},"attestation_state":"computed","paper":{"title":"Unmute the Patch Tokens: Rethinking Probing in Multi-Label Audio Classification","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.SD","authors_text":"Bernhard Sick, Christoph Scholz, Houtan Ghaffari, Ilyass Moummad, Lukas Miklautz, Lukas Rauch, Ren\\'e Heinrich","submitted_at":"2025-09-29T15:11:18Z","abstract_excerpt":"Although probing frozen models has become a standard evaluation paradigm, self-supervised learning in audio defaults to fine-tuning when pursuing state-of-the-art on AudioSet. A key reason is that global pooling creates an information bottleneck causing linear probes to misrepresent the embedding quality: The $\\texttt{cls}$-token discards crucial token information about dispersed, localized events in audio. This weakness is rooted in the mismatch between the pretraining objective (globally) and the downstream task (localized). Across a comprehensive benchmark of 13 datasets and 6 spectrogram-b"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2509.24901","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.SD","submitted_at":"2025-09-29T15:11:18Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"11b7167bb7ceb4508658aff6a86d628a76096df915c2e2b8fc925a483662c65b","abstract_canon_sha256":"d53f67580022387b3f875cf76f48996a4bca58a0302d5c151ce0dfd28b06974a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-01T02:03:27.385797Z","signature_b64":"4KeiTTxIoHQGXd9d9a1nILyNw6bOxm8+o1IlTYwamhnLIARBGGx5MlMMtndCf+I9kkXvCdaAduwtCfgHFNOSDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9b4914f7728b7a9f802a2a876e5bf55d24c382f9ae46aea0b3e26090c475a8c7","last_reissued_at":"2026-06-01T02:03:27.384627Z","signature_status":"signed_v1","first_computed_at":"2026-06-01T02:03:27.384627Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Unmute the Patch Tokens: Rethinking Probing in Multi-Label Audio Classification","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.SD","authors_text":"Bernhard Sick, Christoph Scholz, Houtan Ghaffari, Ilyass Moummad, Lukas Miklautz, Lukas Rauch, Ren\\'e Heinrich","submitted_at":"2025-09-29T15:11:18Z","abstract_excerpt":"Although probing frozen models has become a standard evaluation paradigm, self-supervised learning in audio defaults to fine-tuning when pursuing state-of-the-art on AudioSet. A key reason is that global pooling creates an information bottleneck causing linear probes to misrepresent the embedding quality: The $\\texttt{cls}$-token discards crucial token information about dispersed, localized events in audio. This weakness is rooted in the mismatch between the pretraining objective (globally) and the downstream task (localized). Across a comprehensive benchmark of 13 datasets and 6 spectrogram-b"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.24901","kind":"arxiv","version":4},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2509.24901/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2509.24901","created_at":"2026-06-01T02:03:27.384812+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.24901v4","created_at":"2026-06-01T02:03:27.384812+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.24901","created_at":"2026-06-01T02:03:27.384812+00:00"},{"alias_kind":"pith_short_12","alias_value":"TNERJ53SRN5J","created_at":"2026-06-01T02:03:27.384812+00:00"},{"alias_kind":"pith_short_16","alias_value":"TNERJ53SRN5J7ABK","created_at":"2026-06-01T02:03:27.384812+00:00"},{"alias_kind":"pith_short_8","alias_value":"TNERJ53S","created_at":"2026-06-01T02:03:27.384812+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.14231","citing_title":"AudioMosaic: Contrastive Masked Audio Representation Learning","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU","json":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU.json","graph_json":"https://pith.science/api/pith-number/TNERJ53SRN5J7ABKFKDW4W7VLU/graph.json","events_json":"https://pith.science/api/pith-number/TNERJ53SRN5J7ABKFKDW4W7VLU/events.json","paper":"https://pith.science/paper/TNERJ53S"},"agent_actions":{"view_html":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU","download_json":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU.json","view_paper":"https://pith.science/paper/TNERJ53S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.24901&json=true","fetch_graph":"https://pith.science/api/pith-number/TNERJ53SRN5J7ABKFKDW4W7VLU/graph.json","fetch_events":"https://pith.science/api/pith-number/TNERJ53SRN5J7ABKFKDW4W7VLU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU/action/storage_attestation","attest_author":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU/action/author_attestation","sign_citation":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU/action/citation_signature","submit_replication":"https://pith.science/pith/TNERJ53SRN5J7ABKFKDW4W7VLU/action/replication_record"}},"created_at":"2026-06-01T02:03:27.384812+00:00","updated_at":"2026-06-01T02:03:27.384812+00:00"}