{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:2HPEDC3IFX2TKKR5LSMQ4J6Q2E","short_pith_number":"pith:2HPEDC3I","schema_version":"1.0","canonical_sha256":"d1de418b682df5352a3d5c990e27d0d10f542ecddb96d865b3af78a5ed92b328","source":{"kind":"arxiv","id":"2602.03139","version":2},"attestation_state":"computed","paper":{"title":"Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Kede Ma, Lei Zhang, Ruibin Li, Tianhe Wu","submitted_at":"2026-02-03T05:45:25Z","abstract_excerpt":"Distribution matching distillation (DMD) facilitates few-step image generation by aligning a distilled student with a reference multi-step teacher. In practice, however, optimizing DMD can reduce sample diversity in few-step synthesis, and existing remedies typically rely on perceptual or adversarial regularization, leading to stability and scalability challenges during training. Here, we describe diversity-preserved DMD (DP-DMD), a role-separated distillation method inspired by the complementary roles of early and late denoising steps. Specifically, the first distillation step is trained with"},"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":"2602.03139","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-02-03T05:45:25Z","cross_cats_sorted":[],"title_canon_sha256":"424d1aa780aed064930fe89191efc39f492631e3528712e1e4fa4b50d2d41c80","abstract_canon_sha256":"c64e52a81fd30f844376549fbabac17ba4980f860ac27deb0d0f36f3f5121181"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:06:08.095552Z","signature_b64":"Qk6M/F+ouTqhMAqd1Ghd6c6Hwnm1U21t5Qiyp3CnBOFkFiDI+/6AKmblkjLp6ve+/0BTPxD5gBomiJ8pPly1Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d1de418b682df5352a3d5c990e27d0d10f542ecddb96d865b3af78a5ed92b328","last_reissued_at":"2026-05-20T01:06:08.094903Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:06:08.094903Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Kede Ma, Lei Zhang, Ruibin Li, Tianhe Wu","submitted_at":"2026-02-03T05:45:25Z","abstract_excerpt":"Distribution matching distillation (DMD) facilitates few-step image generation by aligning a distilled student with a reference multi-step teacher. In practice, however, optimizing DMD can reduce sample diversity in few-step synthesis, and existing remedies typically rely on perceptual or adversarial regularization, leading to stability and scalability challenges during training. Here, we describe diversity-preserved DMD (DP-DMD), a role-separated distillation method inspired by the complementary roles of early and late denoising steps. Specifically, the first distillation step is trained with"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.03139","kind":"arxiv","version":2},"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/2602.03139/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":"2602.03139","created_at":"2026-05-20T01:06:08.095005+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.03139v2","created_at":"2026-05-20T01:06:08.095005+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.03139","created_at":"2026-05-20T01:06:08.095005+00:00"},{"alias_kind":"pith_short_12","alias_value":"2HPEDC3IFX2T","created_at":"2026-05-20T01:06:08.095005+00:00"},{"alias_kind":"pith_short_16","alias_value":"2HPEDC3IFX2TKKR5","created_at":"2026-05-20T01:06:08.095005+00:00"},{"alias_kind":"pith_short_8","alias_value":"2HPEDC3I","created_at":"2026-05-20T01:06:08.095005+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2605.11494","citing_title":"STRIDE: Training-Free Diversity Guidance via PCA-Directed Feature Perturbation in Single-Step Diffusion Models","ref_index":38,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11596","citing_title":"HorizonDrive: Self-Corrective Autoregressive World Model for Long-horizon Driving Simulation","ref_index":24,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10730","citing_title":"Qwen-Image-2.0 Technical Report","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2604.10103","citing_title":"Long-Horizon Streaming Video Generation via Hybrid Attention with Decoupled Distillation","ref_index":46,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E","json":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E.json","graph_json":"https://pith.science/api/pith-number/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/graph.json","events_json":"https://pith.science/api/pith-number/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/events.json","paper":"https://pith.science/paper/2HPEDC3I"},"agent_actions":{"view_html":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E","download_json":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E.json","view_paper":"https://pith.science/paper/2HPEDC3I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.03139&json=true","fetch_graph":"https://pith.science/api/pith-number/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/graph.json","fetch_events":"https://pith.science/api/pith-number/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/action/storage_attestation","attest_author":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/action/author_attestation","sign_citation":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/action/citation_signature","submit_replication":"https://pith.science/pith/2HPEDC3IFX2TKKR5LSMQ4J6Q2E/action/replication_record"}},"created_at":"2026-05-20T01:06:08.095005+00:00","updated_at":"2026-05-20T01:06:08.095005+00:00"}