{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:T2OEVGYJ4EERBPHEFGGIXRM2XR","short_pith_number":"pith:T2OEVGYJ","canonical_record":{"source":{"id":"2605.13399","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:54:43Z","cross_cats_sorted":["cs.IT","math.IT"],"title_canon_sha256":"be762bfc797ccf2cd10a08b478f26dc5cab13e9d14ce9fa3a4dba410e09c5e8c","abstract_canon_sha256":"20cb237f9e864eb1045f98640be732ce1f677698723774c2fe46a7c708d8711c"},"schema_version":"1.0"},"canonical_sha256":"9e9c4a9b09e10910bce4298c8bc59abc7c17a36a7144adc1765bbae4e5498037","source":{"kind":"arxiv","id":"2605.13399","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13399","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13399v1","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13399","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"pith_short_12","alias_value":"T2OEVGYJ4EER","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"T2OEVGYJ4EERBPHE","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"T2OEVGYJ","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:T2OEVGYJ4EERBPHEFGGIXRM2XR","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13399","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:54:43Z","cross_cats_sorted":["cs.IT","math.IT"],"title_canon_sha256":"be762bfc797ccf2cd10a08b478f26dc5cab13e9d14ce9fa3a4dba410e09c5e8c","abstract_canon_sha256":"20cb237f9e864eb1045f98640be732ce1f677698723774c2fe46a7c708d8711c"},"schema_version":"1.0"},"canonical_sha256":"9e9c4a9b09e10910bce4298c8bc59abc7c17a36a7144adc1765bbae4e5498037","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:47.611269Z","signature_b64":"eFQ2qo4PxfZS5t74yLvGCPgpddpL7mX3bqaWW1Q0NRliSJZjlwZHCY+eppRsdQz8jVZQPNYQ5e69o+fhpmOrAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9e9c4a9b09e10910bce4298c8bc59abc7c17a36a7144adc1765bbae4e5498037","last_reissued_at":"2026-05-18T02:44:47.610852Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:47.610852Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13399","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":"2026-05-18T02:44:47Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Uw3ik8RwscMZzATGoR9teFEFjmsBIr+H3Vs2JzxLrPikuEcjL/4QA0B8bs9y6GfQfWXz1GAxfmVQrVBfeTi9AA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T23:35:29.619373Z"},"content_sha256":"418aa3739f8bdb7279451041ed408c1e36a5d39cd8977736c08b7506a9cc2887","schema_version":"1.0","event_id":"sha256:418aa3739f8bdb7279451041ed408c1e36a5d39cd8977736c08b7506a9cc2887"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:T2OEVGYJ4EERBPHEFGGIXRM2XR","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"The Diffusion Encoder","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Diffusion models can replace standard encoders in autoencoders when trained alternately with the decoder to align latent estimates.","cross_cats":["cs.IT","math.IT"],"primary_cat":"cs.LG","authors_text":"Akhil Premkumar, Sarah Lucioni","submitted_at":"2026-05-13T11:54:43Z","abstract_excerpt":"We construct a new kind of encoder, leveraging the expressive power of diffusion models. In a traditional variational autoencoder, the encoder and decoder jointly negotiate a latent representation of the input. This is made possible by the reparameterization trick, which simplifies training at the cost of restricting the encoder to a simple family of distributions. Replacing this encoder with a diffusion model requires rethinking how the decoder pressure can be transmitted back to the encoder, given that they tend to update their internal estimates of the latent in opposing directions. We solv"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our method enables more reliable synchronization between encoder and decoder, while preserving the simple and efficient training objective of standard diffusion models.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That an alternating training schedule inspired by EM can transmit decoder gradients back to the diffusion encoder without causing instability or divergence in the latent estimates.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A diffusion model serves as the encoder in an autoencoder when trained alternately with the decoder to resolve opposing update directions while retaining the standard diffusion training objective.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Diffusion models can replace standard encoders in autoencoders when trained alternately with the decoder to align latent estimates.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"128f7d318ca6c20d3da7dd6879eb4d9fc92915ac5b4a68c233ef3c93086f03aa"},"source":{"id":"2605.13399","kind":"arxiv","version":1},"verdict":{"id":"3612a628-346a-442f-9df5-43ee6f3063d1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:10:14.574173Z","strongest_claim":"Our method enables more reliable synchronization between encoder and decoder, while preserving the simple and efficient training objective of standard diffusion models.","one_line_summary":"A diffusion model serves as the encoder in an autoencoder when trained alternately with the decoder to resolve opposing update directions while retaining the standard diffusion training objective.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That an alternating training schedule inspired by EM can transmit decoder gradients back to the diffusion encoder without causing instability or divergence in the latent estimates.","pith_extraction_headline":"Diffusion models can replace standard encoders in autoencoders when trained alternately with the decoder to align latent estimates."},"references":{"count":46,"sample":[{"doi":"","year":2014,"title":"Auto-Encoding Variational Bayes","work_id":"97d95295-30e1-42b4-bbf6-85f0fa4edb44","ref_index":1,"cited_arxiv_id":"1312.6114","is_internal_anchor":true},{"doi":"","year":2025,"title":"DIME: Diffusion-Based Maximum Entropy Reinforcement Learning","work_id":"0f94979f-bd04-441b-8df6-cc4704ae91d4","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Soft Actor-Critic: Off- Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor","work_id":"a06f66e8-9d77-46f9-9234-c296b2b4b7a0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"Q-Learning with Adjoint Matching","work_id":"9fd1472e-aec6-4390-947c-167f083600f5","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Inference Suboptimality in Variational Autoencoders","work_id":"6cabe7c4-6531-41a8-93cc-62813167a24b","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":46,"snapshot_sha256":"6cd58d0db2d77959a5d59dcc9a2c296a0f5494a16496c5cb42e7df0723e9d83e","internal_anchors":4},"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"},"verdict_id":"3612a628-346a-442f-9df5-43ee6f3063d1"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:47Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"wQTtpntDAI01/w7jatGBlpfSR9jVvEw8swOkntn0ZLAJlFY1W1TWxStNZaRmsy/OUDRpdd8BehhqBuox72xkAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-30T23:35:29.620443Z"},"content_sha256":"0b3c5921353a7da7b70653d48eac2bdc766117dd6536afada47740724571d08d","schema_version":"1.0","event_id":"sha256:0b3c5921353a7da7b70653d48eac2bdc766117dd6536afada47740724571d08d"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/T2OEVGYJ4EERBPHEFGGIXRM2XR/bundle.json","state_url":"https://pith.science/pith/T2OEVGYJ4EERBPHEFGGIXRM2XR/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/T2OEVGYJ4EERBPHEFGGIXRM2XR/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":"2026-05-30T23:35:29Z","links":{"resolver":"https://pith.science/pith/T2OEVGYJ4EERBPHEFGGIXRM2XR","bundle":"https://pith.science/pith/T2OEVGYJ4EERBPHEFGGIXRM2XR/bundle.json","state":"https://pith.science/pith/T2OEVGYJ4EERBPHEFGGIXRM2XR/state.json","well_known_bundle":"https://pith.science/.well-known/pith/T2OEVGYJ4EERBPHEFGGIXRM2XR/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:T2OEVGYJ4EERBPHEFGGIXRM2XR","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":"20cb237f9e864eb1045f98640be732ce1f677698723774c2fe46a7c708d8711c","cross_cats_sorted":["cs.IT","math.IT"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:54:43Z","title_canon_sha256":"be762bfc797ccf2cd10a08b478f26dc5cab13e9d14ce9fa3a4dba410e09c5e8c"},"schema_version":"1.0","source":{"id":"2605.13399","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13399","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13399v1","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13399","created_at":"2026-05-18T02:44:47Z"},{"alias_kind":"pith_short_12","alias_value":"T2OEVGYJ4EER","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"T2OEVGYJ4EERBPHE","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"T2OEVGYJ","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:0b3c5921353a7da7b70653d48eac2bdc766117dd6536afada47740724571d08d","target":"graph","created_at":"2026-05-18T02:44:47Z","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":"Our method enables more reliable synchronization between encoder and decoder, while preserving the simple and efficient training objective of standard diffusion models."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That an alternating training schedule inspired by EM can transmit decoder gradients back to the diffusion encoder without causing instability or divergence in the latent estimates."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"A diffusion model serves as the encoder in an autoencoder when trained alternately with the decoder to resolve opposing update directions while retaining the standard diffusion training objective."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Diffusion models can replace standard encoders in autoencoders when trained alternately with the decoder to align latent estimates."}],"snapshot_sha256":"128f7d318ca6c20d3da7dd6879eb4d9fc92915ac5b4a68c233ef3c93086f03aa"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"We construct a new kind of encoder, leveraging the expressive power of diffusion models. In a traditional variational autoencoder, the encoder and decoder jointly negotiate a latent representation of the input. This is made possible by the reparameterization trick, which simplifies training at the cost of restricting the encoder to a simple family of distributions. Replacing this encoder with a diffusion model requires rethinking how the decoder pressure can be transmitted back to the encoder, given that they tend to update their internal estimates of the latent in opposing directions. We solv","authors_text":"Akhil Premkumar, Sarah Lucioni","cross_cats":["cs.IT","math.IT"],"headline":"Diffusion models can replace standard encoders in autoencoders when trained alternately with the decoder to align latent estimates.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:54:43Z","title":"The Diffusion Encoder"},"references":{"count":46,"internal_anchors":4,"resolved_work":46,"sample":[{"cited_arxiv_id":"1312.6114","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Auto-Encoding Variational Bayes","work_id":"97d95295-30e1-42b4-bbf6-85f0fa4edb44","year":2014},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"DIME: Diffusion-Based Maximum Entropy Reinforcement Learning","work_id":"0f94979f-bd04-441b-8df6-cc4704ae91d4","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Soft Actor-Critic: Off- Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor","work_id":"a06f66e8-9d77-46f9-9234-c296b2b4b7a0","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Q-Learning with Adjoint Matching","work_id":"9fd1472e-aec6-4390-947c-167f083600f5","year":2026},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Inference Suboptimality in Variational Autoencoders","work_id":"6cabe7c4-6531-41a8-93cc-62813167a24b","year":2018}],"snapshot_sha256":"6cd58d0db2d77959a5d59dcc9a2c296a0f5494a16496c5cb42e7df0723e9d83e"},"source":{"id":"2605.13399","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T20:10:14.574173Z","id":"3612a628-346a-442f-9df5-43ee6f3063d1","model_set":{"reader":"grok-4.3"},"one_line_summary":"A diffusion model serves as the encoder in an autoencoder when trained alternately with the decoder to resolve opposing update directions while retaining the standard diffusion training objective.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Diffusion models can replace standard encoders in autoencoders when trained alternately with the decoder to align latent estimates.","strongest_claim":"Our method enables more reliable synchronization between encoder and decoder, while preserving the simple and efficient training objective of standard diffusion models.","weakest_assumption":"That an alternating training schedule inspired by EM can transmit decoder gradients back to the diffusion encoder without causing instability or divergence in the latent estimates."}},"verdict_id":"3612a628-346a-442f-9df5-43ee6f3063d1"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:418aa3739f8bdb7279451041ed408c1e36a5d39cd8977736c08b7506a9cc2887","target":"record","created_at":"2026-05-18T02:44:47Z","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":"20cb237f9e864eb1045f98640be732ce1f677698723774c2fe46a7c708d8711c","cross_cats_sorted":["cs.IT","math.IT"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T11:54:43Z","title_canon_sha256":"be762bfc797ccf2cd10a08b478f26dc5cab13e9d14ce9fa3a4dba410e09c5e8c"},"schema_version":"1.0","source":{"id":"2605.13399","kind":"arxiv","version":1}},"canonical_sha256":"9e9c4a9b09e10910bce4298c8bc59abc7c17a36a7144adc1765bbae4e5498037","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"9e9c4a9b09e10910bce4298c8bc59abc7c17a36a7144adc1765bbae4e5498037","first_computed_at":"2026-05-18T02:44:47.610852Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:47.610852Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"eFQ2qo4PxfZS5t74yLvGCPgpddpL7mX3bqaWW1Q0NRliSJZjlwZHCY+eppRsdQz8jVZQPNYQ5e69o+fhpmOrAg==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:47.611269Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13399","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:418aa3739f8bdb7279451041ed408c1e36a5d39cd8977736c08b7506a9cc2887","sha256:0b3c5921353a7da7b70653d48eac2bdc766117dd6536afada47740724571d08d"],"state_sha256":"11875bcdb29e7d20a30ba817f38af3ffd8ab36e8bd1ee2a9114703e33151da55"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"swvbTXn0+nUxDGQaKDyoi8ki3CEkyPTSJogIV7nTQgzWoCNR6ZQ4GZXLNrKtTZ+7iGiBOKEmCKYAOlQ+2mdGDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-30T23:35:29.624931Z","bundle_sha256":"2b89d0410f2bdd9756068e5f17f631276de1e26ef32ae920320d2f65aab1f655"}}