{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:36RQD25OJ62ZTK4M4FKYOEVVSN","short_pith_number":"pith:36RQD25O","schema_version":"1.0","canonical_sha256":"dfa301ebae4fb599ab8ce1558712b5935dfad0526ed8c241eaa77639729fc26b","source":{"kind":"arxiv","id":"2305.16958","version":2},"attestation_state":"computed","paper":{"title":"MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"David Rosenberg, Mark Dredze, Mohit Bansal, Ozan Irsoy, Shijie Wu, Shiyue Zhang, Steven Lu","submitted_at":"2023-05-26T14:14:51Z","abstract_excerpt":"Autoregressive language models are trained by minimizing the cross-entropy of the model distribution Q relative to the data distribution P -- that is, minimizing the forward cross-entropy, which is equivalent to maximum likelihood estimation (MLE). We have observed that models trained in this way may \"over-generalize\", in the sense that they produce non-human-like text. Moreover, we believe that reverse cross-entropy, i.e., the cross-entropy of P relative to Q, is a better reflection of how a human would evaluate text generated by a model. Hence, we propose learning with MixCE, an objective th"},"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":"2305.16958","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-05-26T14:14:51Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"b7cb069c45549099864305f99713a0b3d740c92cefeb268d9915411db4efab42","abstract_canon_sha256":"d4ff8c789f0a7c9ba6e1a3594b721ea730b61cee8dc272f53cc5af9c63833d21"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:23:09.220095Z","signature_b64":"+zFL6NCf5TCSW1ATzflf+2eSxxReA/a8xc0iTsPLkTBWpCRtNb3VQHNyH6oHr6OtyldlLdZcpclTHAQDVasZDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dfa301ebae4fb599ab8ce1558712b5935dfad0526ed8c241eaa77639729fc26b","last_reissued_at":"2026-07-05T08:23:09.219689Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:23:09.219689Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"David Rosenberg, Mark Dredze, Mohit Bansal, Ozan Irsoy, Shijie Wu, Shiyue Zhang, Steven Lu","submitted_at":"2023-05-26T14:14:51Z","abstract_excerpt":"Autoregressive language models are trained by minimizing the cross-entropy of the model distribution Q relative to the data distribution P -- that is, minimizing the forward cross-entropy, which is equivalent to maximum likelihood estimation (MLE). We have observed that models trained in this way may \"over-generalize\", in the sense that they produce non-human-like text. Moreover, we believe that reverse cross-entropy, i.e., the cross-entropy of P relative to Q, is a better reflection of how a human would evaluate text generated by a model. Hence, we propose learning with MixCE, an objective th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.16958","kind":"arxiv","version":2},"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/2305.16958/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":"2305.16958","created_at":"2026-07-05T08:23:09.219747+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.16958v2","created_at":"2026-07-05T08:23:09.219747+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.16958","created_at":"2026-07-05T08:23:09.219747+00:00"},{"alias_kind":"pith_short_12","alias_value":"36RQD25OJ62Z","created_at":"2026-07-05T08:23:09.219747+00:00"},{"alias_kind":"pith_short_16","alias_value":"36RQD25OJ62ZTK4M","created_at":"2026-07-05T08:23:09.219747+00:00"},{"alias_kind":"pith_short_8","alias_value":"36RQD25O","created_at":"2026-07-05T08:23:09.219747+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN","json":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN.json","graph_json":"https://pith.science/api/pith-number/36RQD25OJ62ZTK4M4FKYOEVVSN/graph.json","events_json":"https://pith.science/api/pith-number/36RQD25OJ62ZTK4M4FKYOEVVSN/events.json","paper":"https://pith.science/paper/36RQD25O"},"agent_actions":{"view_html":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN","download_json":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN.json","view_paper":"https://pith.science/paper/36RQD25O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.16958&json=true","fetch_graph":"https://pith.science/api/pith-number/36RQD25OJ62ZTK4M4FKYOEVVSN/graph.json","fetch_events":"https://pith.science/api/pith-number/36RQD25OJ62ZTK4M4FKYOEVVSN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN/action/storage_attestation","attest_author":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN/action/author_attestation","sign_citation":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN/action/citation_signature","submit_replication":"https://pith.science/pith/36RQD25OJ62ZTK4M4FKYOEVVSN/action/replication_record"}},"created_at":"2026-07-05T08:23:09.219747+00:00","updated_at":"2026-07-05T08:23:09.219747+00:00"}