{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YAGX4P3DUXM6PI4AUQPZVK3RJN","short_pith_number":"pith:YAGX4P3D","schema_version":"1.0","canonical_sha256":"c00d7e3f63a5d9e7a380a41f9aab714b5a24fe0e07e8d323ffe54ad3284b5067","source":{"kind":"arxiv","id":"2605.16949","version":1},"attestation_state":"computed","paper":{"title":"Beyond Point-Wise Matching: Structural Representation Alignment for Accelerating Diffusion Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Structural alignment of relational geometry in features accelerates Diffusion Transformer training and improves sample quality.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Houqiang Li, Litong Gong, Shaodong Xu, Tiezheng Ge, Wengang Zhou, Zexian Li, Zhendong Wang","submitted_at":"2026-05-16T12:01:04Z","abstract_excerpt":"Recent advances in Diffusion Transformers (DiTs) demonstrate that aligning noisy latent states with well-trained semantic features-as pioneered by Representation Alignment (REPA)-can substantially accelerate training and improve generation fidelity. Subsequent analysis(e.g., iREPA) suggests that these gains arise primarily from transferring spatial structure contained in pre-trained vision representations. However, mostly existing alignment methods employ point-wise matching objectives or rely on implicit architectural tweaks, which fail to explicitly model the spatial relational geometry inhe"},"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":"2605.16949","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-16T12:01:04Z","cross_cats_sorted":[],"title_canon_sha256":"beb3178ecbdb85edbc2ca2e09f733f4131143ee911495293b4d2ced987e8325a","abstract_canon_sha256":"d7ad830213bbfaa4f34c7ae1d28792b25fe8353fbc53f2e56c106155751bbe6b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:03:32.460545Z","signature_b64":"Wxzff1Nx2/cxcPvpI2qPL1i27slGs3kz4QqAfBt8pG49MAsmj/QByAZcBlnYyabAygkRsPiLRA/NEXPaapwqAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c00d7e3f63a5d9e7a380a41f9aab714b5a24fe0e07e8d323ffe54ad3284b5067","last_reissued_at":"2026-05-20T00:03:32.459882Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:03:32.459882Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond Point-Wise Matching: Structural Representation Alignment for Accelerating Diffusion Transformers","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Structural alignment of relational geometry in features accelerates Diffusion Transformer training and improves sample quality.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Houqiang Li, Litong Gong, Shaodong Xu, Tiezheng Ge, Wengang Zhou, Zexian Li, Zhendong Wang","submitted_at":"2026-05-16T12:01:04Z","abstract_excerpt":"Recent advances in Diffusion Transformers (DiTs) demonstrate that aligning noisy latent states with well-trained semantic features-as pioneered by Representation Alignment (REPA)-can substantially accelerate training and improve generation fidelity. Subsequent analysis(e.g., iREPA) suggests that these gains arise primarily from transferring spatial structure contained in pre-trained vision representations. However, mostly existing alignment methods employ point-wise matching objectives or rely on implicit architectural tweaks, which fail to explicitly model the spatial relational geometry inhe"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By encouraging the model to internalize holistic spatial layouts and structural correlations from pre-trained features, sREPA achieves faster and more stable convergence, along with improved sample quality, compared to state-of-the-art alignment strategies.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That point-wise matching objectives are insufficient to capture the rich spatial topology of visual representations and that an explicit structural constraint on relational geometry will transfer this topology more effectively.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"sREPA enforces structural consistency in relational geometry of pre-trained vision features to accelerate DiT training and improve generation quality.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Structural alignment of relational geometry in features accelerates Diffusion Transformer training and improves sample quality.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"6074a1f87421b7b67f63e069fbd51d0673c48571aa5b0fd1ecc7bac93b4ca406"},"source":{"id":"2605.16949","kind":"arxiv","version":1},"verdict":{"id":"3750419e-ba72-4662-80d1-0c7618d1de4b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T20:46:10.172474Z","strongest_claim":"By encouraging the model to internalize holistic spatial layouts and structural correlations from pre-trained features, sREPA achieves faster and more stable convergence, along with improved sample quality, compared to state-of-the-art alignment strategies.","one_line_summary":"sREPA enforces structural consistency in relational geometry of pre-trained vision features to accelerate DiT training and improve generation quality.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That point-wise matching objectives are insufficient to capture the rich spatial topology of visual representations and that an explicit structural constraint on relational geometry will transfer this topology more effectively.","pith_extraction_headline":"Structural alignment of relational geometry in features accelerates Diffusion Transformer training and improves sample quality."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16949/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:01:19.108306Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T20:50:51.242078Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T19:52:11.328303Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T18:41:56.240017Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.323329Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"a1eeb7243df3fafeb44e77a775d4b0e46d7f11697b183df3d07b44f29887b792"},"references":{"count":46,"sample":[{"doi":"","year":2023,"title":"Self-supervised learning from images with a joint-embedding predictive architecture","work_id":"4336ef84-2bc6-4a25-a713-cefd1f8eea15","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Video generation models as world simulators.OpenAI Blog, 1(8):1, 2024","work_id":"9da2c350-7ff5-4ba2-98a7-c641ea0ab2bd","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"An empirical study of training self-supervised vision transformers","work_id":"2574a999-f6c9-45a0-9375-dc9a96e1fa2e","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"Imagenet: A large-scale hierarchical image database","work_id":"cc291e4b-478b-4e79-ab37-d782c8e1888e","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Diffusion models beat gans on image synthesis.Advances in neural information processing systems, 34:8780–8794","work_id":"3bd99a68-1f13-405d-b9a9-8934c9454ce4","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":46,"snapshot_sha256":"077b547bbda3f5dcbf192e199394f4e88b67c879311b46405cc91609d52d5937","internal_anchors":14},"formal_canon":{"evidence_count":3,"snapshot_sha256":"ecd89ca87dc7c3c24c19364c825f47394b85ae2ff8add1b264d01f7c07c9f998"},"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":"2605.16949","created_at":"2026-05-20T00:03:32.459970+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.16949v1","created_at":"2026-05-20T00:03:32.459970+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.16949","created_at":"2026-05-20T00:03:32.459970+00:00"},{"alias_kind":"pith_short_12","alias_value":"YAGX4P3DUXM6","created_at":"2026-05-20T00:03:32.459970+00:00"},{"alias_kind":"pith_short_16","alias_value":"YAGX4P3DUXM6PI4A","created_at":"2026-05-20T00:03:32.459970+00:00"},{"alias_kind":"pith_short_8","alias_value":"YAGX4P3D","created_at":"2026-05-20T00:03:32.459970+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":3,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN","json":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN.json","graph_json":"https://pith.science/api/pith-number/YAGX4P3DUXM6PI4AUQPZVK3RJN/graph.json","events_json":"https://pith.science/api/pith-number/YAGX4P3DUXM6PI4AUQPZVK3RJN/events.json","paper":"https://pith.science/paper/YAGX4P3D"},"agent_actions":{"view_html":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN","download_json":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN.json","view_paper":"https://pith.science/paper/YAGX4P3D","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.16949&json=true","fetch_graph":"https://pith.science/api/pith-number/YAGX4P3DUXM6PI4AUQPZVK3RJN/graph.json","fetch_events":"https://pith.science/api/pith-number/YAGX4P3DUXM6PI4AUQPZVK3RJN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN/action/storage_attestation","attest_author":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN/action/author_attestation","sign_citation":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN/action/citation_signature","submit_replication":"https://pith.science/pith/YAGX4P3DUXM6PI4AUQPZVK3RJN/action/replication_record"}},"created_at":"2026-05-20T00:03:32.459970+00:00","updated_at":"2026-05-20T00:03:32.459970+00:00"}