{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:UV3QR62BY2IXQMHTAXPHGLVELO","short_pith_number":"pith:UV3QR62B","schema_version":"1.0","canonical_sha256":"a57708fb41c6917830f305de732ea45ba376aa1f1aab348cba50d9312f3c38f8","source":{"kind":"arxiv","id":"2208.02646","version":4},"attestation_state":"computed","paper":{"title":"DropKey","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bonan Li, Congying Han, Luoqi Liu, Tiande Guo, Xiangjian Jiang, Xuecheng Nie, Yinhan Hu","submitted_at":"2022-08-04T13:24:04Z","abstract_excerpt":"In this paper, we focus on analyzing and improving the dropout technique for self-attention layers of Vision Transformer, which is important while surprisingly ignored by prior works. In particular, we conduct researches on three core questions: First, what to drop in self-attention layers? Different from dropping attention weights in literature, we propose to move dropout operations forward ahead of attention matrix calculation and set the Key as the dropout unit, yielding a novel dropout-before-softmax scheme. We theoretically verify that this scheme helps keep both regularization and probab"},"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":"2208.02646","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2022-08-04T13:24:04Z","cross_cats_sorted":[],"title_canon_sha256":"fe53e80102b4dcb9a9bfecb70ef1b10131f492750da126621660eab16beabd03","abstract_canon_sha256":"c2258db02a2e79880421b070bf090ff9e8bead83631459b1dec51b89faae6300"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:59:38.658252Z","signature_b64":"/Ujaz5RX4PTsPjj8XpGmXzH1GPV/FtcfVMXscSki1925/v7Puxc7+Iu1HvXyYxd2jIrR6awOLvGdotmcxUFLBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a57708fb41c6917830f305de732ea45ba376aa1f1aab348cba50d9312f3c38f8","last_reissued_at":"2026-07-05T05:59:38.657797Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:59:38.657797Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DropKey","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bonan Li, Congying Han, Luoqi Liu, Tiande Guo, Xiangjian Jiang, Xuecheng Nie, Yinhan Hu","submitted_at":"2022-08-04T13:24:04Z","abstract_excerpt":"In this paper, we focus on analyzing and improving the dropout technique for self-attention layers of Vision Transformer, which is important while surprisingly ignored by prior works. In particular, we conduct researches on three core questions: First, what to drop in self-attention layers? Different from dropping attention weights in literature, we propose to move dropout operations forward ahead of attention matrix calculation and set the Key as the dropout unit, yielding a novel dropout-before-softmax scheme. We theoretically verify that this scheme helps keep both regularization and probab"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2208.02646","kind":"arxiv","version":4},"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/2208.02646/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":"2208.02646","created_at":"2026-07-05T05:59:38.657855+00:00"},{"alias_kind":"arxiv_version","alias_value":"2208.02646v4","created_at":"2026-07-05T05:59:38.657855+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2208.02646","created_at":"2026-07-05T05:59:38.657855+00:00"},{"alias_kind":"pith_short_12","alias_value":"UV3QR62BY2IX","created_at":"2026-07-05T05:59:38.657855+00:00"},{"alias_kind":"pith_short_16","alias_value":"UV3QR62BY2IXQMHT","created_at":"2026-07-05T05:59:38.657855+00:00"},{"alias_kind":"pith_short_8","alias_value":"UV3QR62B","created_at":"2026-07-05T05:59:38.657855+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.03221","citing_title":"Neural Inhibition Improves Dynamic Routing and Mixture of Experts","ref_index":22,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO","json":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO.json","graph_json":"https://pith.science/api/pith-number/UV3QR62BY2IXQMHTAXPHGLVELO/graph.json","events_json":"https://pith.science/api/pith-number/UV3QR62BY2IXQMHTAXPHGLVELO/events.json","paper":"https://pith.science/paper/UV3QR62B"},"agent_actions":{"view_html":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO","download_json":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO.json","view_paper":"https://pith.science/paper/UV3QR62B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2208.02646&json=true","fetch_graph":"https://pith.science/api/pith-number/UV3QR62BY2IXQMHTAXPHGLVELO/graph.json","fetch_events":"https://pith.science/api/pith-number/UV3QR62BY2IXQMHTAXPHGLVELO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO/action/storage_attestation","attest_author":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO/action/author_attestation","sign_citation":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO/action/citation_signature","submit_replication":"https://pith.science/pith/UV3QR62BY2IXQMHTAXPHGLVELO/action/replication_record"}},"created_at":"2026-07-05T05:59:38.657855+00:00","updated_at":"2026-07-05T05:59:38.657855+00:00"}