{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:N6AO33LZWQNIKSVMIRLEDNT67B","short_pith_number":"pith:N6AO33LZ","schema_version":"1.0","canonical_sha256":"6f80eded79b41a854aac445641b67ef8641d24172c21e3dbab18eccc18fda828","source":{"kind":"arxiv","id":"2009.08092","version":2},"attestation_state":"computed","paper":{"title":"Distributional Generalization: A New Kind of Generalization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.NE","math.ST","stat.ML","stat.TH"],"primary_cat":"cs.LG","authors_text":"Preetum Nakkiran, Yamini Bansal","submitted_at":"2020-09-17T06:26:17Z","abstract_excerpt":"We introduce a new notion of generalization -- Distributional Generalization -- which roughly states that outputs of a classifier at train and test time are close *as distributions*, as opposed to close in just their average error. For example, if we mislabel 30% of dogs as cats in the train set of CIFAR-10, then a ResNet trained to interpolation will in fact mislabel roughly 30% of dogs as cats on the *test set* as well, while leaving other classes unaffected. This behavior is not captured by classical generalization, which would only consider the average error and not the distribution of err"},"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":"2009.08092","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-09-17T06:26:17Z","cross_cats_sorted":["cs.NE","math.ST","stat.ML","stat.TH"],"title_canon_sha256":"13a956a08c938304f1b3646e815c27429663a26bb5b5118ec3e17ac844e9c974","abstract_canon_sha256":"15b37bd79a5dfb8b6bb38dc8f1a71a48f2b308272cbc640163aaf4dc786d559a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:43:16.213718Z","signature_b64":"wMRux8DUgwT7SqKX+KFGBviOPl5Go/Oav9z3f5iZquuROOke+Dn5yTQhjif13AnZp9Mdkq9MAGEU1IruxlIIDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6f80eded79b41a854aac445641b67ef8641d24172c21e3dbab18eccc18fda828","last_reissued_at":"2026-07-05T01:43:16.213288Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:43:16.213288Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Distributional Generalization: A New Kind of Generalization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.NE","math.ST","stat.ML","stat.TH"],"primary_cat":"cs.LG","authors_text":"Preetum Nakkiran, Yamini Bansal","submitted_at":"2020-09-17T06:26:17Z","abstract_excerpt":"We introduce a new notion of generalization -- Distributional Generalization -- which roughly states that outputs of a classifier at train and test time are close *as distributions*, as opposed to close in just their average error. For example, if we mislabel 30% of dogs as cats in the train set of CIFAR-10, then a ResNet trained to interpolation will in fact mislabel roughly 30% of dogs as cats on the *test set* as well, while leaving other classes unaffected. This behavior is not captured by classical generalization, which would only consider the average error and not the distribution of err"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2009.08092","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/2009.08092/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":"2009.08092","created_at":"2026-07-05T01:43:16.213363+00:00"},{"alias_kind":"arxiv_version","alias_value":"2009.08092v2","created_at":"2026-07-05T01:43:16.213363+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2009.08092","created_at":"2026-07-05T01:43:16.213363+00:00"},{"alias_kind":"pith_short_12","alias_value":"N6AO33LZWQNI","created_at":"2026-07-05T01:43:16.213363+00:00"},{"alias_kind":"pith_short_16","alias_value":"N6AO33LZWQNIKSVM","created_at":"2026-07-05T01:43:16.213363+00:00"},{"alias_kind":"pith_short_8","alias_value":"N6AO33LZ","created_at":"2026-07-05T01:43:16.213363+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.17767","citing_title":"Feature Learning in Linear-Width Two-Layer Networks: Two vs. One Step of Gradient Descent","ref_index":86,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17767","citing_title":"Feature Learning in Linear-Width Two-Layer Networks: Two vs. One Step of Gradient Descent","ref_index":86,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B","json":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B.json","graph_json":"https://pith.science/api/pith-number/N6AO33LZWQNIKSVMIRLEDNT67B/graph.json","events_json":"https://pith.science/api/pith-number/N6AO33LZWQNIKSVMIRLEDNT67B/events.json","paper":"https://pith.science/paper/N6AO33LZ"},"agent_actions":{"view_html":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B","download_json":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B.json","view_paper":"https://pith.science/paper/N6AO33LZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2009.08092&json=true","fetch_graph":"https://pith.science/api/pith-number/N6AO33LZWQNIKSVMIRLEDNT67B/graph.json","fetch_events":"https://pith.science/api/pith-number/N6AO33LZWQNIKSVMIRLEDNT67B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B/action/storage_attestation","attest_author":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B/action/author_attestation","sign_citation":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B/action/citation_signature","submit_replication":"https://pith.science/pith/N6AO33LZWQNIKSVMIRLEDNT67B/action/replication_record"}},"created_at":"2026-07-05T01:43:16.213363+00:00","updated_at":"2026-07-05T01:43:16.213363+00:00"}