{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2021:FG6NN5OFUW4YDWSRN6NW62UAU2","short_pith_number":"pith:FG6NN5OF","canonical_record":{"source":{"id":"2101.02388","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2021-01-07T06:12:28Z","cross_cats_sorted":[],"title_canon_sha256":"77b47e855f9bdf2d5e24f76a15a93a8c1d4f746862ac7e734cc8330101aa1828","abstract_canon_sha256":"bedeb1003efae2e366f4270f01eb5faf462b9df1938b129a4a2f3654a8cf6a9d"},"schema_version":"1.0"},"canonical_sha256":"29bcd6f5c5a5b981da516f9b6f6a80a686b9c6ca167911ff8b9cee3050ec0efc","source":{"kind":"arxiv","id":"2101.02388","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2101.02388","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2101.02388v1","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2101.02388","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"FG6NN5OFUW4Y","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"FG6NN5OFUW4YDWSR","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"FG6NN5OF","created_at":"2026-05-18T12:33:33Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2021:FG6NN5OFUW4YDWSRN6NW62UAU2","target":"record","payload":{"canonical_record":{"source":{"id":"2101.02388","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2021-01-07T06:12:28Z","cross_cats_sorted":[],"title_canon_sha256":"77b47e855f9bdf2d5e24f76a15a93a8c1d4f746862ac7e734cc8330101aa1828","abstract_canon_sha256":"bedeb1003efae2e366f4270f01eb5faf462b9df1938b129a4a2f3654a8cf6a9d"},"schema_version":"1.0"},"canonical_sha256":"29bcd6f5c5a5b981da516f9b6f6a80a686b9c6ca167911ff8b9cee3050ec0efc","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:15.200933Z","signature_b64":"jvO/QgVszZEqjV/Mis4dZKZe/noHeUL5Pu/Rw4haOKIHBCDLlaN20POF/CLgo/s4/blpBXNNEXttX4MvwY3HBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"29bcd6f5c5a5b981da516f9b6f6a80a686b9c6ca167911ff8b9cee3050ec0efc","last_reissued_at":"2026-05-17T23:38:15.200262Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:15.200262Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2101.02388","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"6M+/WdBaLNE2BSq+ocI5vTSTw89NBewdeeFsPRrWdHZD/DSXR6zjmpNO/QQlkDFTOXr51ODCitHEkE8jbMKhDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T18:34:56.170149Z"},"content_sha256":"c02538d2d09a7062246fef8893bf44adc9004f880d1713fe6b4909d7024346d9","schema_version":"1.0","event_id":"sha256:c02538d2d09a7062246fef8893bf44adc9004f880d1713fe6b4909d7024346d9"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2021:FG6NN5OFUW4YDWSRN6NW62UAU2","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Knowledge Distillation in Iterative Generative Models for Improved Sampling Speed","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"Knowledge distillation turns a multi-step denoising model into a fast single-step generator matching GAN quality.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eric Luhman, Troy Luhman","submitted_at":"2021-01-07T06:12:28Z","abstract_excerpt":"Iterative generative models, such as noise conditional score networks and denoising diffusion probabilistic models, produce high quality samples by gradually denoising an initial noise vector. However, their denoising process has many steps, making them 2-3 orders of magnitude slower than other generative models such as GANs and VAEs. In this paper, we establish a novel connection between knowledge distillation and image generation with a technique that distills a multi-step denoising process into a single step, resulting in a sampling speed similar to other single-step generative models. Our "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We establish a novel connection between knowledge distillation and image generation with a technique that distills a multi-step denoising process into a single step, resulting in a sampling speed similar to other single-step generative models. Our Denoising Student generates high quality samples comparable to GANs on the CIFAR-10 and CelebA datasets, without adversarial training.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That a single forward pass through the student can faithfully approximate the distribution produced by the full multi-step teacher denoising trajectory without requiring additional regularization or architectural changes that would re-introduce iteration.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Denoising Student distills the multi-step denoising process of score-based and diffusion models into a single forward pass, matching GAN sampling speed while producing comparable sample quality on CIFAR-10, CelebA, and 256x256 LSUN.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Knowledge distillation turns a multi-step denoising model into a fast single-step generator matching GAN quality.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"9cced91d0e81e197191b11b4b134105e396f06c57187400f42ded7d401ec7bce"},"source":{"id":"2101.02388","kind":"arxiv","version":1},"verdict":{"id":"b0ca5e01-22a2-4584-9900-cf292c868caf","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T03:52:10.568856Z","strongest_claim":"We establish a novel connection between knowledge distillation and image generation with a technique that distills a multi-step denoising process into a single step, resulting in a sampling speed similar to other single-step generative models. Our Denoising Student generates high quality samples comparable to GANs on the CIFAR-10 and CelebA datasets, without adversarial training.","one_line_summary":"Denoising Student distills the multi-step denoising process of score-based and diffusion models into a single forward pass, matching GAN sampling speed while producing comparable sample quality on CIFAR-10, CelebA, and 256x256 LSUN.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That a single forward pass through the student can faithfully approximate the distribution produced by the full multi-step teacher denoising trajectory without requiring additional regularization or architectural changes that would re-introduce iteration.","pith_extraction_headline":"Knowledge distillation turns a multi-step denoising model into a fast single-step generator matching GAN quality."},"references":{"count":51,"sample":[{"doi":"","year":2019,"title":"Large Scale GAN Training for High Fidelity Natural Image Synthesis","work_id":"f64eb098-b61e-40fa-98c5-1c99418f0f29","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1145/1150402.1150464","year":2006,"title":"Model Compres- sion","work_id":"45d23c68-c1e5-4371-82db-b3d8081e0808","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Distilling Knowledge from Ensembles of Neural Networks for Speech Recognition","work_id":"2c8242f0-75d4-46ed-8c4f-596e07dab5d9","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Very deep vaes generalize autoregressive models and can outperform them on images","work_id":"214ff54a-9ea1-46bc-94a9-8daf75f1d2b9","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Implicit generation and modeling with energy based models","work_id":"abc4382e-5bd6-45e6-873f-f41c6963874f","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":51,"snapshot_sha256":"05b4b1529cbcd134144f81638375a4a1aa24164f19d311f7c7673e61532b3bef","internal_anchors":14},"formal_canon":{"evidence_count":2,"snapshot_sha256":"b7f927a45151ff356d5a78ead900ba3d835afc9a0401758c54195b8a1bac77b6"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"b0ca5e01-22a2-4584-9900-cf292c868caf"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:38:15Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gKya9lNvgnijcK7XZ8NHhbSlPwcz68T1YHieCy3X0i02HH5y9ashUrYXlqC1G1noVGOSwZupgpiA5AWqDiKLDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T18:34:56.171262Z"},"content_sha256":"d2324ea1ea94eb1e12bf44db5158860ebea28b299ca7cab0e0e36c959c2f1bba","schema_version":"1.0","event_id":"sha256:d2324ea1ea94eb1e12bf44db5158860ebea28b299ca7cab0e0e36c959c2f1bba"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/FG6NN5OFUW4YDWSRN6NW62UAU2/bundle.json","state_url":"https://pith.science/pith/FG6NN5OFUW4YDWSRN6NW62UAU2/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/FG6NN5OFUW4YDWSRN6NW62UAU2/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-08T18:34:56Z","links":{"resolver":"https://pith.science/pith/FG6NN5OFUW4YDWSRN6NW62UAU2","bundle":"https://pith.science/pith/FG6NN5OFUW4YDWSRN6NW62UAU2/bundle.json","state":"https://pith.science/pith/FG6NN5OFUW4YDWSRN6NW62UAU2/state.json","well_known_bundle":"https://pith.science/.well-known/pith/FG6NN5OFUW4YDWSRN6NW62UAU2/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2021:FG6NN5OFUW4YDWSRN6NW62UAU2","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"bedeb1003efae2e366f4270f01eb5faf462b9df1938b129a4a2f3654a8cf6a9d","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2021-01-07T06:12:28Z","title_canon_sha256":"77b47e855f9bdf2d5e24f76a15a93a8c1d4f746862ac7e734cc8330101aa1828"},"schema_version":"1.0","source":{"id":"2101.02388","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2101.02388","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"arxiv_version","alias_value":"2101.02388v1","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2101.02388","created_at":"2026-05-17T23:38:15Z"},{"alias_kind":"pith_short_12","alias_value":"FG6NN5OFUW4Y","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"FG6NN5OFUW4YDWSR","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"FG6NN5OF","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:d2324ea1ea94eb1e12bf44db5158860ebea28b299ca7cab0e0e36c959c2f1bba","target":"graph","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"We establish a novel connection between knowledge distillation and image generation with a technique that distills a multi-step denoising process into a single step, resulting in a sampling speed similar to other single-step generative models. Our Denoising Student generates high quality samples comparable to GANs on the CIFAR-10 and CelebA datasets, without adversarial training."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That a single forward pass through the student can faithfully approximate the distribution produced by the full multi-step teacher denoising trajectory without requiring additional regularization or architectural changes that would re-introduce iteration."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Denoising Student distills the multi-step denoising process of score-based and diffusion models into a single forward pass, matching GAN sampling speed while producing comparable sample quality on CIFAR-10, CelebA, and 256x256 LSUN."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Knowledge distillation turns a multi-step denoising model into a fast single-step generator matching GAN quality."}],"snapshot_sha256":"9cced91d0e81e197191b11b4b134105e396f06c57187400f42ded7d401ec7bce"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"b7f927a45151ff356d5a78ead900ba3d835afc9a0401758c54195b8a1bac77b6"},"paper":{"abstract_excerpt":"Iterative generative models, such as noise conditional score networks and denoising diffusion probabilistic models, produce high quality samples by gradually denoising an initial noise vector. However, their denoising process has many steps, making them 2-3 orders of magnitude slower than other generative models such as GANs and VAEs. In this paper, we establish a novel connection between knowledge distillation and image generation with a technique that distills a multi-step denoising process into a single step, resulting in a sampling speed similar to other single-step generative models. Our ","authors_text":"Eric Luhman, Troy Luhman","cross_cats":[],"headline":"Knowledge distillation turns a multi-step denoising model into a fast single-step generator matching GAN quality.","license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2021-01-07T06:12:28Z","title":"Knowledge Distillation in Iterative Generative Models for Improved Sampling Speed"},"references":{"count":51,"internal_anchors":14,"resolved_work":51,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Large Scale GAN Training for High Fidelity Natural Image Synthesis","work_id":"f64eb098-b61e-40fa-98c5-1c99418f0f29","year":2019},{"cited_arxiv_id":"","doi":"10.1145/1150402.1150464","is_internal_anchor":false,"ref_index":2,"title":"Model Compres- sion","work_id":"45d23c68-c1e5-4371-82db-b3d8081e0808","year":2006},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Distilling Knowledge from Ensembles of Neural Networks for Speech Recognition","work_id":"2c8242f0-75d4-46ed-8c4f-596e07dab5d9","year":2016},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Very deep vaes generalize autoregressive models and can outperform them on images","work_id":"214ff54a-9ea1-46bc-94a9-8daf75f1d2b9","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Implicit generation and modeling with energy based models","work_id":"abc4382e-5bd6-45e6-873f-f41c6963874f","year":2019}],"snapshot_sha256":"05b4b1529cbcd134144f81638375a4a1aa24164f19d311f7c7673e61532b3bef"},"source":{"id":"2101.02388","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-17T03:52:10.568856Z","id":"b0ca5e01-22a2-4584-9900-cf292c868caf","model_set":{"reader":"grok-4.3"},"one_line_summary":"Denoising Student distills the multi-step denoising process of score-based and diffusion models into a single forward pass, matching GAN sampling speed while producing comparable sample quality on CIFAR-10, CelebA, and 256x256 LSUN.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Knowledge distillation turns a multi-step denoising model into a fast single-step generator matching GAN quality.","strongest_claim":"We establish a novel connection between knowledge distillation and image generation with a technique that distills a multi-step denoising process into a single step, resulting in a sampling speed similar to other single-step generative models. Our Denoising Student generates high quality samples comparable to GANs on the CIFAR-10 and CelebA datasets, without adversarial training.","weakest_assumption":"That a single forward pass through the student can faithfully approximate the distribution produced by the full multi-step teacher denoising trajectory without requiring additional regularization or architectural changes that would re-introduce iteration."}},"verdict_id":"b0ca5e01-22a2-4584-9900-cf292c868caf"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c02538d2d09a7062246fef8893bf44adc9004f880d1713fe6b4909d7024346d9","target":"record","created_at":"2026-05-17T23:38:15Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"bedeb1003efae2e366f4270f01eb5faf462b9df1938b129a4a2f3654a8cf6a9d","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2021-01-07T06:12:28Z","title_canon_sha256":"77b47e855f9bdf2d5e24f76a15a93a8c1d4f746862ac7e734cc8330101aa1828"},"schema_version":"1.0","source":{"id":"2101.02388","kind":"arxiv","version":1}},"canonical_sha256":"29bcd6f5c5a5b981da516f9b6f6a80a686b9c6ca167911ff8b9cee3050ec0efc","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"29bcd6f5c5a5b981da516f9b6f6a80a686b9c6ca167911ff8b9cee3050ec0efc","first_computed_at":"2026-05-17T23:38:15.200262Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:15.200262Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"jvO/QgVszZEqjV/Mis4dZKZe/noHeUL5Pu/Rw4haOKIHBCDLlaN20POF/CLgo/s4/blpBXNNEXttX4MvwY3HBA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:15.200933Z","signed_message":"canonical_sha256_bytes"},"source_id":"2101.02388","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c02538d2d09a7062246fef8893bf44adc9004f880d1713fe6b4909d7024346d9","sha256:d2324ea1ea94eb1e12bf44db5158860ebea28b299ca7cab0e0e36c959c2f1bba"],"state_sha256":"3bb0fa486ea1d87042d084069189398d0ec0b2c7450cbb20d98f9219f4458e15"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"GBL0DwwUu6ru2wMi4YjnQ0fHgKDU5TDHpwyy29Y6RIH/jQcGA+tlOiXuvW9ppXiygDYg66MLMBJ7r9Swz3nEDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-08T18:34:56.176178Z","bundle_sha256":"fd05e7405ae3dbb8e2cc8c5d07f2c8fe6420e4e9cc95b43172e2ae7f3a062412"}}