{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:WK42L2EZKBOJRRC7TF4LQ7M6ZH","short_pith_number":"pith:WK42L2EZ","schema_version":"1.0","canonical_sha256":"b2b9a5e899505c98c45f9978b87d9ec9c810a46d98bbf2690e91106a8b1ef861","source":{"kind":"arxiv","id":"2208.11970","version":1},"attestation_state":"computed","paper":{"title":"Understanding Diffusion Models: A Unified Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Calvin Luo","submitted_at":"2022-08-25T09:55:25Z","abstract_excerpt":"Diffusion models have shown incredible capabilities as generative models; indeed, they power the current state-of-the-art models on text-conditioned image generation such as Imagen and DALL-E 2. In this work we review, demystify, and unify the understanding of diffusion models across both variational and score-based perspectives. We first derive Variational Diffusion Models (VDM) as a special case of a Markovian Hierarchical Variational Autoencoder, where three key assumptions enable tractable computation and scalable optimization of the ELBO. We then prove that optimizing a VDM boils down to "},"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":"2208.11970","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-08-25T09:55:25Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"37f33795818c2218fa3a916f31c7273d7bab0ced8d9de544634e526146581f58","abstract_canon_sha256":"a20f9e70968997bff887ea40bd1b367aed486697a2cec913fdacf13981c269da"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:51:28.186162Z","signature_b64":"l2Q9HlqXSG9XebXHOSSsaY1Xb+tvl8wRs0jfOTZdwsknS+HkW0lQ67uXtzpHLVuP352OlbUM+qA+9wk1Le02CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b2b9a5e899505c98c45f9978b87d9ec9c810a46d98bbf2690e91106a8b1ef861","last_reissued_at":"2026-07-05T04:51:28.185805Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:51:28.185805Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Diffusion Models: A Unified Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Calvin Luo","submitted_at":"2022-08-25T09:55:25Z","abstract_excerpt":"Diffusion models have shown incredible capabilities as generative models; indeed, they power the current state-of-the-art models on text-conditioned image generation such as Imagen and DALL-E 2. In this work we review, demystify, and unify the understanding of diffusion models across both variational and score-based perspectives. We first derive Variational Diffusion Models (VDM) as a special case of a Markovian Hierarchical Variational Autoencoder, where three key assumptions enable tractable computation and scalable optimization of the ELBO. We then prove that optimizing a VDM boils down to "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2208.11970","kind":"arxiv","version":1},"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/2208.11970/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":"2208.11970","created_at":"2026-07-05T04:51:28.185859+00:00"},{"alias_kind":"arxiv_version","alias_value":"2208.11970v1","created_at":"2026-07-05T04:51:28.185859+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2208.11970","created_at":"2026-07-05T04:51:28.185859+00:00"},{"alias_kind":"pith_short_12","alias_value":"WK42L2EZKBOJ","created_at":"2026-07-05T04:51:28.185859+00:00"},{"alias_kind":"pith_short_16","alias_value":"WK42L2EZKBOJRRC7","created_at":"2026-07-05T04:51:28.185859+00:00"},{"alias_kind":"pith_short_8","alias_value":"WK42L2EZ","created_at":"2026-07-05T04:51:28.185859+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":29,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08335","citing_title":"Bayesian Experimental Design via Score Matching","ref_index":127,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24982","citing_title":"Latent Block-Diffusion Temporal Point Processes: A Semi-Autoregressive Framework for Asynchronous Event Sequence Generation","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20416","citing_title":"On the Redundancy of Timestep Embeddings in Diffusion Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01645","citing_title":"Self-Regulating Annealing in Heavy-Tailed Diffusion Models","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27637","citing_title":"AI-Generated Image Recognition via Fusion of CNNs and Vision Transformers","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29807","citing_title":"Rethinking Forgery Attacks on Semantic Watermarks in Black-Box Settings: A Geometric Distortion Perspective","ref_index":67,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30267","citing_title":"Pathway variability, coat stiffening and mechanical adaptation during clathrin-mediated endocytosis","ref_index":114,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28264","citing_title":"A Two-Step Ensemble Score Filter for Data Assimilation in Partially Observed Systems","ref_index":90,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00673","citing_title":"T-CLIP: Enabling Thermal Perception for Contrastive Language-Image Pretraining","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07220","citing_title":"On the Robustness of Distribution Support under Diffusion Guidance","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2311.04938","citing_title":"Improved DDIM Sampling with Moment Matching Gaussian Mixtures","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2405.16772","citing_title":"Balancing User Preferences by Social Networks: A Condition-Guided Social Recommendation Model for Mitigating Popularity Bias","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2412.03134","citing_title":"A Probabilistic Formulation of Offset Noise in Diffusion Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2412.08637","citing_title":"DMin: Scalable Training Data Influence Estimation for Diffusion Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17804","citing_title":"GenTS: A Comprehensive Benchmark Library for Generative Time Series Models","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18325","citing_title":"Mixture-of-Experts Diffusion Models for Adaptive Massive MIMO Channel Estimation via Variational Bayesian Inference","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2505.19519","citing_title":"Preserve and Personalize: Personalized Text-to-Image Diffusion Models without Distributional Drift","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2506.21107","citing_title":"Doloris: Dual Conditional Diffusion Implicit Bridges with Sparsity Masking Strategy for Unpaired Single-Cell Perturbation Estimation","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2507.06121","citing_title":"Brownian Bridge Diffusion for Sequential Recommendation","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2509.25003","citing_title":"Score-based Membership Inference on Diffusion Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03303","citing_title":"Downscaling weather forecasts from Low- to High-Resolution with Diffusion Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27911","citing_title":"Physical Foundation Models: Fixed hardware implementations of large-scale neural networks","ref_index":95,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10790","citing_title":"Elucidating Representation Degradation Problem in Diffusion Model Training","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08915","citing_title":"Physics-Informed Neural PDE Solvers via Spatio-Temporal MeanFlow","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17952","citing_title":"Causal inference for social network formation","ref_index":151,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH","json":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH.json","graph_json":"https://pith.science/api/pith-number/WK42L2EZKBOJRRC7TF4LQ7M6ZH/graph.json","events_json":"https://pith.science/api/pith-number/WK42L2EZKBOJRRC7TF4LQ7M6ZH/events.json","paper":"https://pith.science/paper/WK42L2EZ"},"agent_actions":{"view_html":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH","download_json":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH.json","view_paper":"https://pith.science/paper/WK42L2EZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2208.11970&json=true","fetch_graph":"https://pith.science/api/pith-number/WK42L2EZKBOJRRC7TF4LQ7M6ZH/graph.json","fetch_events":"https://pith.science/api/pith-number/WK42L2EZKBOJRRC7TF4LQ7M6ZH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH/action/storage_attestation","attest_author":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH/action/author_attestation","sign_citation":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH/action/citation_signature","submit_replication":"https://pith.science/pith/WK42L2EZKBOJRRC7TF4LQ7M6ZH/action/replication_record"}},"created_at":"2026-07-05T04:51:28.185859+00:00","updated_at":"2026-07-05T04:51:28.185859+00:00"}