{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:FK7UVWJZASMGVRODOKXV4DTWLU","short_pith_number":"pith:FK7UVWJZ","schema_version":"1.0","canonical_sha256":"2abf4ad93904986ac5c372af5e0e765d31928a91304d02f66ee875631cbb19c3","source":{"kind":"arxiv","id":"2006.07322","version":5},"attestation_state":"computed","paper":{"title":"Evaluation of Neural Architectures Trained with Square Loss vs Cross-Entropy in Classification Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Like Hui, Mikhail Belkin","submitted_at":"2020-06-12T17:00:49Z","abstract_excerpt":"Modern neural architectures for classification tasks are trained using the cross-entropy loss, which is widely believed to be empirically superior to the square loss. In this work we provide evidence indicating that this belief may not be well-founded. We explore several major neural architectures and a range of standard benchmark datasets for NLP, automatic speech recognition (ASR) and computer vision tasks to show that these architectures, with the same hyper-parameter settings as reported in the literature, perform comparably or better when trained with the square loss, even after equalizin"},"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":"2006.07322","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-06-12T17:00:49Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"0be3c89ef24f4e65b48f539d8d51da484705f64b6a0c4a68d7ff28e5ed5a97d7","abstract_canon_sha256":"9dfc1def78f745629c35d5b323cef35915e2baf25c83011d6de98153ad759d84"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:24:54.719067Z","signature_b64":"cJwDoSiX2iQfg0d1h+qmtTg0ho8q8ZFWsP7W2LKGWzujidApOCo3zwnij43pEwyPznbcW8qaAAtmm+b7BTxRAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2abf4ad93904986ac5c372af5e0e765d31928a91304d02f66ee875631cbb19c3","last_reissued_at":"2026-07-05T03:24:54.718562Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:24:54.718562Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Evaluation of Neural Architectures Trained with Square Loss vs Cross-Entropy in Classification Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Like Hui, Mikhail Belkin","submitted_at":"2020-06-12T17:00:49Z","abstract_excerpt":"Modern neural architectures for classification tasks are trained using the cross-entropy loss, which is widely believed to be empirically superior to the square loss. In this work we provide evidence indicating that this belief may not be well-founded. We explore several major neural architectures and a range of standard benchmark datasets for NLP, automatic speech recognition (ASR) and computer vision tasks to show that these architectures, with the same hyper-parameter settings as reported in the literature, perform comparably or better when trained with the square loss, even after equalizin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2006.07322","kind":"arxiv","version":5},"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/2006.07322/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":"2006.07322","created_at":"2026-07-05T03:24:54.718631+00:00"},{"alias_kind":"arxiv_version","alias_value":"2006.07322v5","created_at":"2026-07-05T03:24:54.718631+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2006.07322","created_at":"2026-07-05T03:24:54.718631+00:00"},{"alias_kind":"pith_short_12","alias_value":"FK7UVWJZASMG","created_at":"2026-07-05T03:24:54.718631+00:00"},{"alias_kind":"pith_short_16","alias_value":"FK7UVWJZASMGVROD","created_at":"2026-07-05T03:24:54.718631+00:00"},{"alias_kind":"pith_short_8","alias_value":"FK7UVWJZ","created_at":"2026-07-05T03:24:54.718631+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22068","citing_title":"Alternate loss functions and regression models that achieve robustness to outliers by modulating the learning rate","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31581","citing_title":"Robustness of neural networks to random noise perturbations of their inputs","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06689","citing_title":"Generative Cross-Entropy: A Strictly Proper Loss for Data-Efficient Classification","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06689","citing_title":"Generative Cross-Entropy: A Strictly Proper Loss for Data-Efficient Classification","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU","json":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU.json","graph_json":"https://pith.science/api/pith-number/FK7UVWJZASMGVRODOKXV4DTWLU/graph.json","events_json":"https://pith.science/api/pith-number/FK7UVWJZASMGVRODOKXV4DTWLU/events.json","paper":"https://pith.science/paper/FK7UVWJZ"},"agent_actions":{"view_html":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU","download_json":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU.json","view_paper":"https://pith.science/paper/FK7UVWJZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2006.07322&json=true","fetch_graph":"https://pith.science/api/pith-number/FK7UVWJZASMGVRODOKXV4DTWLU/graph.json","fetch_events":"https://pith.science/api/pith-number/FK7UVWJZASMGVRODOKXV4DTWLU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU/action/storage_attestation","attest_author":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU/action/author_attestation","sign_citation":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU/action/citation_signature","submit_replication":"https://pith.science/pith/FK7UVWJZASMGVRODOKXV4DTWLU/action/replication_record"}},"created_at":"2026-07-05T03:24:54.718631+00:00","updated_at":"2026-07-05T03:24:54.718631+00:00"}