{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:CSUTGYMODOYM3J3EEUQZH6OBUH","short_pith_number":"pith:CSUTGYMO","schema_version":"1.0","canonical_sha256":"14a933618e1bb0cda764252193f9c1a1ea8b5ee5703c2cb4d933cd7de611242f","source":{"kind":"arxiv","id":"2004.05529","version":1},"attestation_state":"computed","paper":{"title":"Gradients as Features for Deep Representation Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Fangzhou Mu, Yingyu Liang, Yin Li","submitted_at":"2020-04-12T02:57:28Z","abstract_excerpt":"We address the challenging problem of deep representation learning--the efficient adaption of a pre-trained deep network to different tasks. Specifically, we propose to explore gradient-based features. These features are gradients of the model parameters with respect to a task-specific loss given an input sample. Our key innovation is the design of a linear model that incorporates both gradient and activation of the pre-trained network. We show that our model provides a local linear approximation to an underlying deep model, and discuss important theoretical insights. Moreover, we present an e"},"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":"2004.05529","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2020-04-12T02:57:28Z","cross_cats_sorted":["cs.CV","stat.ML"],"title_canon_sha256":"b68dbb1738dc7fb0b581964b73cde865e5271d6637b31f9beab59a1554be8626","abstract_canon_sha256":"31356bc5269606b014b3fa967084787e2b052cc238ed59c4543a403c54556809"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:54:38.070785Z","signature_b64":"WzJkcs+pTey6y4ThUhyI/pMxt2Z8hhz9KiXkMUHibjM9aJFfL9PGmkw5AHicZKfJ+HjPfxY6O7ntp5GG4RaVDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"14a933618e1bb0cda764252193f9c1a1ea8b5ee5703c2cb4d933cd7de611242f","last_reissued_at":"2026-07-05T00:54:38.070343Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:54:38.070343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradients as Features for Deep Representation Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Fangzhou Mu, Yingyu Liang, Yin Li","submitted_at":"2020-04-12T02:57:28Z","abstract_excerpt":"We address the challenging problem of deep representation learning--the efficient adaption of a pre-trained deep network to different tasks. Specifically, we propose to explore gradient-based features. These features are gradients of the model parameters with respect to a task-specific loss given an input sample. Our key innovation is the design of a linear model that incorporates both gradient and activation of the pre-trained network. We show that our model provides a local linear approximation to an underlying deep model, and discuss important theoretical insights. Moreover, we present an e"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2004.05529","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/2004.05529/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":"2004.05529","created_at":"2026-07-05T00:54:38.070409+00:00"},{"alias_kind":"arxiv_version","alias_value":"2004.05529v1","created_at":"2026-07-05T00:54:38.070409+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2004.05529","created_at":"2026-07-05T00:54:38.070409+00:00"},{"alias_kind":"pith_short_12","alias_value":"CSUTGYMODOYM","created_at":"2026-07-05T00:54:38.070409+00:00"},{"alias_kind":"pith_short_16","alias_value":"CSUTGYMODOYM3J3E","created_at":"2026-07-05T00:54:38.070409+00:00"},{"alias_kind":"pith_short_8","alias_value":"CSUTGYMO","created_at":"2026-07-05T00:54:38.070409+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.26999","citing_title":"Compositional Meta-Learning for Mitigating Task Heterogeneity in Physics-Informed Neural Networks","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH","json":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH.json","graph_json":"https://pith.science/api/pith-number/CSUTGYMODOYM3J3EEUQZH6OBUH/graph.json","events_json":"https://pith.science/api/pith-number/CSUTGYMODOYM3J3EEUQZH6OBUH/events.json","paper":"https://pith.science/paper/CSUTGYMO"},"agent_actions":{"view_html":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH","download_json":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH.json","view_paper":"https://pith.science/paper/CSUTGYMO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2004.05529&json=true","fetch_graph":"https://pith.science/api/pith-number/CSUTGYMODOYM3J3EEUQZH6OBUH/graph.json","fetch_events":"https://pith.science/api/pith-number/CSUTGYMODOYM3J3EEUQZH6OBUH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH/action/storage_attestation","attest_author":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH/action/author_attestation","sign_citation":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH/action/citation_signature","submit_replication":"https://pith.science/pith/CSUTGYMODOYM3J3EEUQZH6OBUH/action/replication_record"}},"created_at":"2026-07-05T00:54:38.070409+00:00","updated_at":"2026-07-05T00:54:38.070409+00:00"}