{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:CIGLBWOGLEREBSN3WDO326PEOZ","short_pith_number":"pith:CIGLBWOG","schema_version":"1.0","canonical_sha256":"120cb0d9c6592240c9bbb0ddbd79e4765887b550c0fc3084220747dd2cbf7a53","source":{"kind":"arxiv","id":"2312.04709","version":1},"attestation_state":"computed","paper":{"title":"How to guess a gradient","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NE"],"primary_cat":"cs.LG","authors_text":"Brian Cheung, Jonathan Ragan-Kelley, Joshua B. Tenenbaum, Kartik Chandra, Stella X. Yu, Tomaso A. Poggio, Utkarsh Singhal","submitted_at":"2023-12-07T21:40:44Z","abstract_excerpt":"How much can you say about the gradient of a neural network without computing a loss or knowing the label? This may sound like a strange question: surely the answer is \"very little.\" However, in this paper, we show that gradients are more structured than previously thought. Gradients lie in a predictable low-dimensional subspace which depends on the network architecture and incoming features. Exploiting this structure can significantly improve gradient-free optimization schemes based on directional derivatives, which have struggled to scale beyond small networks trained on toy datasets. We stu"},"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":"2312.04709","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-12-07T21:40:44Z","cross_cats_sorted":["cs.NE"],"title_canon_sha256":"aaaf3110ca74102952dcb94d2f4b3412768158d48d0c237b056aa340cb757404","abstract_canon_sha256":"2403660db249c2d2403af997dbc57c7c64ea0bc8d8e2e2c457f374f2ef71d0ac"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:21:50.119613Z","signature_b64":"kZegq9M/qLnJStSjLomsNX9s5M+mCE/9dMoDtLEcL49wJjMmzrwbI8c00x6qtMsXV70LAVKLskrEdQX0ssTjBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"120cb0d9c6592240c9bbb0ddbd79e4765887b550c0fc3084220747dd2cbf7a53","last_reissued_at":"2026-07-05T07:21:50.119174Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:21:50.119174Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How to guess a gradient","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NE"],"primary_cat":"cs.LG","authors_text":"Brian Cheung, Jonathan Ragan-Kelley, Joshua B. Tenenbaum, Kartik Chandra, Stella X. Yu, Tomaso A. Poggio, Utkarsh Singhal","submitted_at":"2023-12-07T21:40:44Z","abstract_excerpt":"How much can you say about the gradient of a neural network without computing a loss or knowing the label? This may sound like a strange question: surely the answer is \"very little.\" However, in this paper, we show that gradients are more structured than previously thought. Gradients lie in a predictable low-dimensional subspace which depends on the network architecture and incoming features. Exploiting this structure can significantly improve gradient-free optimization schemes based on directional derivatives, which have struggled to scale beyond small networks trained on toy datasets. We stu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.04709","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/2312.04709/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":"2312.04709","created_at":"2026-07-05T07:21:50.119230+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.04709v1","created_at":"2026-07-05T07:21:50.119230+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.04709","created_at":"2026-07-05T07:21:50.119230+00:00"},{"alias_kind":"pith_short_12","alias_value":"CIGLBWOGLERE","created_at":"2026-07-05T07:21:50.119230+00:00"},{"alias_kind":"pith_short_16","alias_value":"CIGLBWOGLEREBSN3","created_at":"2026-07-05T07:21:50.119230+00:00"},{"alias_kind":"pith_short_8","alias_value":"CIGLBWOG","created_at":"2026-07-05T07:21:50.119230+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09734","citing_title":"Adaptive directional gradients for parameterised quantum circuits","ref_index":59,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ","json":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ.json","graph_json":"https://pith.science/api/pith-number/CIGLBWOGLEREBSN3WDO326PEOZ/graph.json","events_json":"https://pith.science/api/pith-number/CIGLBWOGLEREBSN3WDO326PEOZ/events.json","paper":"https://pith.science/paper/CIGLBWOG"},"agent_actions":{"view_html":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ","download_json":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ.json","view_paper":"https://pith.science/paper/CIGLBWOG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.04709&json=true","fetch_graph":"https://pith.science/api/pith-number/CIGLBWOGLEREBSN3WDO326PEOZ/graph.json","fetch_events":"https://pith.science/api/pith-number/CIGLBWOGLEREBSN3WDO326PEOZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ/action/storage_attestation","attest_author":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ/action/author_attestation","sign_citation":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ/action/citation_signature","submit_replication":"https://pith.science/pith/CIGLBWOGLEREBSN3WDO326PEOZ/action/replication_record"}},"created_at":"2026-07-05T07:21:50.119230+00:00","updated_at":"2026-07-05T07:21:50.119230+00:00"}