{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HG6IZUGSFT5BA73UDPMEC6MJYY","short_pith_number":"pith:HG6IZUGS","schema_version":"1.0","canonical_sha256":"39bc8cd0d22cfa107f741bd8417989c6332121cbfa107fab6e3815d8ed17f91f","source":{"kind":"arxiv","id":"2605.15284","version":1},"attestation_state":"computed","paper":{"title":"Tadpole: Autoencoders as Foundation Models for 3D PDEs with Online Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Autoencoders pre-trained on single-channel spatial crops of synthetic 3D PDE data transfer to dynamics learning and generation across diverse physical systems.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Benjamin Holzschuh, Felix Koehler, Nils Thuerey, Qiang Liu","submitted_at":"2026-05-14T18:00:58Z","abstract_excerpt":"We introduce Tadpole, a novel foundation model for three-dimensional partial differential equations (PDEs) that addresses key challenges in transferability, scalability to high dimensionality, and multi-functionality. Tadpole is pre-trained as an autoencoder on synthetic 3D PDE data generated by an efficient online data-generation framework. This enables large-scale, diverse training without storage or I/O overhead, demonstrated by scaling to an equivalent of hundreds of terabytes of training data. By autoencoding single-channel spatial crops, Tadpole learns rich and transferable representatio"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.15284","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T18:00:58Z","cross_cats_sorted":[],"title_canon_sha256":"4e04b9d4ace2e745657aa227e72f10f217718025de921d2d36881d080af5ccaf","abstract_canon_sha256":"b0b29426463e6016b65e28b6c0164c4bd9e3746f853020d469bfe758bcd7ed75"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:50.620676Z","signature_b64":"6Gs+uC/NEOjel54mf35LZZbqZxzP5zc+L1C8fTI0ccE2vgu6PQIlHRdUJQU6NmR1H3C0w3tz/rey+zJ8L1aMAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"39bc8cd0d22cfa107f741bd8417989c6332121cbfa107fab6e3815d8ed17f91f","last_reissued_at":"2026-05-20T00:00:50.619893Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:50.619893Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tadpole: Autoencoders as Foundation Models for 3D PDEs with Online Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Autoencoders pre-trained on single-channel spatial crops of synthetic 3D PDE data transfer to dynamics learning and generation across diverse physical systems.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Benjamin Holzschuh, Felix Koehler, Nils Thuerey, Qiang Liu","submitted_at":"2026-05-14T18:00:58Z","abstract_excerpt":"We introduce Tadpole, a novel foundation model for three-dimensional partial differential equations (PDEs) that addresses key challenges in transferability, scalability to high dimensionality, and multi-functionality. Tadpole is pre-trained as an autoencoder on synthetic 3D PDE data generated by an efficient online data-generation framework. This enables large-scale, diverse training without storage or I/O overhead, demonstrated by scaling to an equivalent of hundreds of terabytes of training data. By autoencoding single-channel spatial crops, Tadpole learns rich and transferable representatio"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Although pre-trained solely as an autoencoder on single-channel spatial crops, Tadpole learns rich and transferable representations across heterogeneous physical systems with varying numbers of state variables and spatial resolutions and can be efficiently applied for multiple downstream tasks including dynamics learning via a novel parameter-efficient fine-tuning strategy.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That representations learned by autoencoding single-channel spatial crops from online-generated synthetic 3D PDE data will transfer effectively to temporal dynamics modeling and generative tasks across different physical systems without requiring large amounts of task-specific data or architecture changes.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Tadpole is a pre-trained autoencoder foundation model for 3D PDEs that learns transferable representations from online-generated data and supports efficient fine-tuning for dynamics prediction and other tasks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Autoencoders pre-trained on single-channel spatial crops of synthetic 3D PDE data transfer to dynamics learning and generation across diverse physical systems.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d05ca65334233fbaff9b055b9ac47316f7bd5529b75693d262221dc088ea00b3"},"source":{"id":"2605.15284","kind":"arxiv","version":1},"verdict":{"id":"813f7463-74a9-4c68-a61b-c36db67a0ba0","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T16:36:58.524944Z","strongest_claim":"Although pre-trained solely as an autoencoder on single-channel spatial crops, Tadpole learns rich and transferable representations across heterogeneous physical systems with varying numbers of state variables and spatial resolutions and can be efficiently applied for multiple downstream tasks including dynamics learning via a novel parameter-efficient fine-tuning strategy.","one_line_summary":"Tadpole is a pre-trained autoencoder foundation model for 3D PDEs that learns transferable representations from online-generated data and supports efficient fine-tuning for dynamics prediction and other tasks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That representations learned by autoencoding single-channel spatial crops from online-generated synthetic 3D PDE data will transfer effectively to temporal dynamics modeling and generative tasks across different physical systems without requiring large amounts of task-specific data or architecture changes.","pith_extraction_headline":"Autoencoders pre-trained on single-channel spatial crops of synthetic 3D PDE data transfer to dynamics learning and generation across diverse physical systems."},"integrity":{"clean":false,"summary":{"advisory":1,"critical":2,"by_detector":{"doi_compliance":{"total":3,"advisory":1,"critical":2,"informational":0}},"informational":0},"endpoint":"/pith/2605.15284/integrity.json","findings":[{"note":"Identifier '10.1016/j.neucom.2021.04' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","detector":"doi_compliance","severity":"critical","ref_index":3,"audited_at":"2026-05-19T16:45:43.939515Z","detected_doi":"10.1016/j.neucom.2021.04","finding_type":"unresolvable_identifier","verdict_class":"cross_source","detected_arxiv_id":null},{"note":"Identifier '10.1006/jcph.2002' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","detector":"doi_compliance","severity":"critical","ref_index":1,"audited_at":"2026-05-19T16:45:43.939515Z","detected_doi":"10.1006/jcph.2002","finding_type":"unresolvable_identifier","verdict_class":"cross_source","detected_arxiv_id":null},{"note":"DOI in the printed bibliography is fragmented by whitespace or line breaks. A longer candidate (10.1103/PhysRevE.111.035304.URL) was visible in the surrounding text but could not be confirmed against doi.org as printed.","detector":"doi_compliance","severity":"advisory","ref_index":7,"audited_at":"2026-05-19T16:45:43.939515Z","detected_doi":"10.1103/PhysRevE.111.035304.URL","finding_type":"recoverable_identifier","verdict_class":"incontrovertible","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T17:01:18.320283Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T16:45:43.939515Z","status":"completed","version":"1.0.0","findings_count":3},{"name":"claim_evidence","ran_at":"2026-05-19T14:41:54.247016Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.790002Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"025755f5e831184ecf85bb41e77be9819df1947deb1d5f6b35dd197b69f8f268"},"references":{"count":14,"sample":[{"doi":"10.1006/jcph.2002","year":2002,"title":"URL https://proceedings.mlr.press/ v235/chen24n.html. Cox, S. and Matthews, P. Exponential Time Dif- ferencing for Stiff Systems. Journal of Compu- tational Physics , 176(2):430–455, 2002. ISSN 0021-9","work_id":"da0bdf42-cfd5-4591-86f7-df7083a9ba8b","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1038/s42256-023-00626-4","year":2023,"title":"2023 , month = mar, publisher =","work_id":"d1e59f5c-8198-43f9-b077-ef9576f2103e","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1016/j.neucom.2021.04","year":2021,"title":"doi: https://doi.org/10.1016/j.neucom.2021.04","work_id":"344af4a4-9e1e-4007-bfdb-f3001dc38741","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Holzschuh, B., Liu, Q., Kohl, G., and Thuerey, N","work_id":"fe367cfe-40f5-44bd-8d73-4544821d2e74","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1137/24m1636071","year":2024,"title":"URL https:// epubs.siam.org/doi/10.1137/24M1636071","work_id":"4c458799-2784-4ac5-bd3f-e8f7b6fc1ad0","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":14,"snapshot_sha256":"abc6c09dbb9c7162e4c2bc1867821f298ab3ce461769e72bdaf5f42047d7389a","internal_anchors":1},"formal_canon":{"evidence_count":2,"snapshot_sha256":"90ed0f54474bfeb63e34e4be711d0c4060b97b626909173fa717f78c5cc5a5a8"},"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":"2605.15284","created_at":"2026-05-20T00:00:50.620047+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15284v1","created_at":"2026-05-20T00:00:50.620047+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15284","created_at":"2026-05-20T00:00:50.620047+00:00"},{"alias_kind":"pith_short_12","alias_value":"HG6IZUGSFT5B","created_at":"2026-05-20T00:00:50.620047+00:00"},{"alias_kind":"pith_short_16","alias_value":"HG6IZUGSFT5BA73U","created_at":"2026-05-20T00:00:50.620047+00:00"},{"alias_kind":"pith_short_8","alias_value":"HG6IZUGS","created_at":"2026-05-20T00:00:50.620047+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY","json":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY.json","graph_json":"https://pith.science/api/pith-number/HG6IZUGSFT5BA73UDPMEC6MJYY/graph.json","events_json":"https://pith.science/api/pith-number/HG6IZUGSFT5BA73UDPMEC6MJYY/events.json","paper":"https://pith.science/paper/HG6IZUGS"},"agent_actions":{"view_html":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY","download_json":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY.json","view_paper":"https://pith.science/paper/HG6IZUGS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15284&json=true","fetch_graph":"https://pith.science/api/pith-number/HG6IZUGSFT5BA73UDPMEC6MJYY/graph.json","fetch_events":"https://pith.science/api/pith-number/HG6IZUGSFT5BA73UDPMEC6MJYY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY/action/storage_attestation","attest_author":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY/action/author_attestation","sign_citation":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY/action/citation_signature","submit_replication":"https://pith.science/pith/HG6IZUGSFT5BA73UDPMEC6MJYY/action/replication_record"}},"created_at":"2026-05-20T00:00:50.620047+00:00","updated_at":"2026-05-20T00:00:50.620047+00:00"}