{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:5XJIXR3UES7OORNW353BAZZYRJ","short_pith_number":"pith:5XJIXR3U","schema_version":"1.0","canonical_sha256":"edd28bc77424bee745b6df761067388a7b90c898ab123027911393cdd3db0f57","source":{"kind":"arxiv","id":"2108.02446","version":3},"attestation_state":"computed","paper":{"title":"Finetuning Pretrained Transformers into Variational Autoencoders","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jihwa Lee, Seongmin Park","submitted_at":"2021-08-05T08:27:26Z","abstract_excerpt":"Text variational autoencoders (VAEs) are notorious for posterior collapse, a phenomenon where the model's decoder learns to ignore signals from the encoder. Because posterior collapse is known to be exacerbated by expressive decoders, Transformers have seen limited adoption as components of text VAEs. Existing studies that incorporate Transformers into text VAEs (Li et al., 2020; Fang et al., 2021) mitigate posterior collapse using massive pretraining, a technique unavailable to most of the research community without extensive computing resources. We present a simple two-phase training scheme "},"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":"2108.02446","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-08-05T08:27:26Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"ec31789f9aeb66a588e0a60f59540412be4ab56c6deefe7d26b725738ecfbe21","abstract_canon_sha256":"fe94c58b18ed5e8b5118ca410e509bd6208dbb6db20a9c854dc8da6a19902e7d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:34:46.004868Z","signature_b64":"VsGhnT+6DpYm5nsck8aF9nf/06soYvUzfORru99bPAepZAHPchUYKyFS73Yf3Z1sEgXLiW5dJ33Z4Lor/vflDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"edd28bc77424bee745b6df761067388a7b90c898ab123027911393cdd3db0f57","last_reissued_at":"2026-07-05T03:34:46.004441Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:34:46.004441Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Finetuning Pretrained Transformers into Variational Autoencoders","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jihwa Lee, Seongmin Park","submitted_at":"2021-08-05T08:27:26Z","abstract_excerpt":"Text variational autoencoders (VAEs) are notorious for posterior collapse, a phenomenon where the model's decoder learns to ignore signals from the encoder. Because posterior collapse is known to be exacerbated by expressive decoders, Transformers have seen limited adoption as components of text VAEs. Existing studies that incorporate Transformers into text VAEs (Li et al., 2020; Fang et al., 2021) mitigate posterior collapse using massive pretraining, a technique unavailable to most of the research community without extensive computing resources. We present a simple two-phase training scheme "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2108.02446","kind":"arxiv","version":3},"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/2108.02446/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":"2108.02446","created_at":"2026-07-05T03:34:46.004498+00:00"},{"alias_kind":"arxiv_version","alias_value":"2108.02446v3","created_at":"2026-07-05T03:34:46.004498+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2108.02446","created_at":"2026-07-05T03:34:46.004498+00:00"},{"alias_kind":"pith_short_12","alias_value":"5XJIXR3UES7O","created_at":"2026-07-05T03:34:46.004498+00:00"},{"alias_kind":"pith_short_16","alias_value":"5XJIXR3UES7OORNW","created_at":"2026-07-05T03:34:46.004498+00:00"},{"alias_kind":"pith_short_8","alias_value":"5XJIXR3U","created_at":"2026-07-05T03:34:46.004498+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2412.20834","citing_title":"Disentangling Preference Representation and Text Generation for Efficient Individual Preference Alignment","ref_index":35,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ","json":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ.json","graph_json":"https://pith.science/api/pith-number/5XJIXR3UES7OORNW353BAZZYRJ/graph.json","events_json":"https://pith.science/api/pith-number/5XJIXR3UES7OORNW353BAZZYRJ/events.json","paper":"https://pith.science/paper/5XJIXR3U"},"agent_actions":{"view_html":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ","download_json":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ.json","view_paper":"https://pith.science/paper/5XJIXR3U","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2108.02446&json=true","fetch_graph":"https://pith.science/api/pith-number/5XJIXR3UES7OORNW353BAZZYRJ/graph.json","fetch_events":"https://pith.science/api/pith-number/5XJIXR3UES7OORNW353BAZZYRJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ/action/storage_attestation","attest_author":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ/action/author_attestation","sign_citation":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ/action/citation_signature","submit_replication":"https://pith.science/pith/5XJIXR3UES7OORNW353BAZZYRJ/action/replication_record"}},"created_at":"2026-07-05T03:34:46.004498+00:00","updated_at":"2026-07-05T03:34:46.004498+00:00"}