{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:7TW4UKQNDAI3BPHPVIJ4JHKYSD","short_pith_number":"pith:7TW4UKQN","schema_version":"1.0","canonical_sha256":"fcedca2a0d1811b0bcefaa13c49d5890c59e653c958d13e24a78912f02202ea6","source":{"kind":"arxiv","id":"2303.09556","version":3},"attestation_state":"computed","paper":{"title":"Efficient Diffusion Training via Min-SNR Weighting Strategy","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Baining Guo, Chen Li, Dong Chen, Han Hu, Jianmin Bao, Shuyang Gu, Tiankai Hang, Xin Geng","submitted_at":"2023-03-16T17:59:56Z","abstract_excerpt":"Denoising diffusion models have been a mainstream approach for image generation, however, training these models often suffers from slow convergence. In this paper, we discovered that the slow convergence is partly due to conflicting optimization directions between timesteps. To address this issue, we treat the diffusion training as a multi-task learning problem, and introduce a simple yet effective approach referred to as Min-SNR-$\\gamma$. This method adapts loss weights of timesteps based on clamped signal-to-noise ratios, which effectively balances the conflicts among timesteps. Our results "},"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":"2303.09556","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2023-03-16T17:59:56Z","cross_cats_sorted":[],"title_canon_sha256":"e7f1c4c9632e3d564b6674179416e6b28954668e3495356146f3563c5e4b49cb","abstract_canon_sha256":"723d584d108d3489651a440a969999f6b0fc67e64ec35d1b20ad1ecbd461fef0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:54:29.896524Z","signature_b64":"U3LwJ8EymZ3cnOHH2ECJTmp2a+Xzu/Q0ibZjPQWV4CXfWSfJtFCE9BPj/EJ84egUsVt8pykq9ja3BtOJL2cxCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fcedca2a0d1811b0bcefaa13c49d5890c59e653c958d13e24a78912f02202ea6","last_reissued_at":"2026-07-05T07:54:29.896022Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:54:29.896022Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Efficient Diffusion Training via Min-SNR Weighting Strategy","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Baining Guo, Chen Li, Dong Chen, Han Hu, Jianmin Bao, Shuyang Gu, Tiankai Hang, Xin Geng","submitted_at":"2023-03-16T17:59:56Z","abstract_excerpt":"Denoising diffusion models have been a mainstream approach for image generation, however, training these models often suffers from slow convergence. In this paper, we discovered that the slow convergence is partly due to conflicting optimization directions between timesteps. To address this issue, we treat the diffusion training as a multi-task learning problem, and introduce a simple yet effective approach referred to as Min-SNR-$\\gamma$. This method adapts loss weights of timesteps based on clamped signal-to-noise ratios, which effectively balances the conflicts among timesteps. Our results "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2303.09556","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/2303.09556/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":"2303.09556","created_at":"2026-07-05T07:54:29.896083+00:00"},{"alias_kind":"arxiv_version","alias_value":"2303.09556v3","created_at":"2026-07-05T07:54:29.896083+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2303.09556","created_at":"2026-07-05T07:54:29.896083+00:00"},{"alias_kind":"pith_short_12","alias_value":"7TW4UKQNDAI3","created_at":"2026-07-05T07:54:29.896083+00:00"},{"alias_kind":"pith_short_16","alias_value":"7TW4UKQNDAI3BPHP","created_at":"2026-07-05T07:54:29.896083+00:00"},{"alias_kind":"pith_short_8","alias_value":"7TW4UKQN","created_at":"2026-07-05T07:54:29.896083+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20722","citing_title":"Multimodal Image Colorization: Quantifying the Impact of Text-Conditioned Guidance on Grayscale-to-Color Translation","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07207","citing_title":"Entropy as a Structural Prior: How a Log-Barrier on DiT Belief Space Drives Musical Diversity and Development","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2511.07571","citing_title":"Forecasting implied volatility surface with generative diffusion models","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2310.15110","citing_title":"Zero123++: a Single Image to Consistent Multi-view Diffusion Base Model","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09168","citing_title":"ELT: Elastic Looped Transformers for Visual Generation","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD","json":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD.json","graph_json":"https://pith.science/api/pith-number/7TW4UKQNDAI3BPHPVIJ4JHKYSD/graph.json","events_json":"https://pith.science/api/pith-number/7TW4UKQNDAI3BPHPVIJ4JHKYSD/events.json","paper":"https://pith.science/paper/7TW4UKQN"},"agent_actions":{"view_html":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD","download_json":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD.json","view_paper":"https://pith.science/paper/7TW4UKQN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2303.09556&json=true","fetch_graph":"https://pith.science/api/pith-number/7TW4UKQNDAI3BPHPVIJ4JHKYSD/graph.json","fetch_events":"https://pith.science/api/pith-number/7TW4UKQNDAI3BPHPVIJ4JHKYSD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD/action/storage_attestation","attest_author":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD/action/author_attestation","sign_citation":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD/action/citation_signature","submit_replication":"https://pith.science/pith/7TW4UKQNDAI3BPHPVIJ4JHKYSD/action/replication_record"}},"created_at":"2026-07-05T07:54:29.896083+00:00","updated_at":"2026-07-05T07:54:29.896083+00:00"}