{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:QRZVECMPSOKANJJI2LRG7AL3BC","short_pith_number":"pith:QRZVECMP","schema_version":"1.0","canonical_sha256":"847352098f939406a528d2e26f817b0893d9e58feba2979edc7efa1e18049d9f","source":{"kind":"arxiv","id":"2502.03029","version":3},"attestation_state":"computed","paper":{"title":"On Zero-Initialized Attention: Optimal Prompt and Gating Factor Estimation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chau Nguyen, Daniel Sonntag, Duy M. H. Nguyen, Huy Nguyen, Mathias Niepert, Minh Le, Nghiem T. Diep, Nhat Ho","submitted_at":"2025-02-05T09:31:27Z","abstract_excerpt":"The LLaMA-Adapter has recently emerged as an efficient fine-tuning technique for LLaMA models, leveraging zero-initialized attention to stabilize training and enhance performance. However, despite its empirical success, the theoretical foundations of zero-initialized attention remain largely unexplored. In this paper, we provide a rigorous theoretical analysis, establishing a connection between zero-initialized attention and mixture-of-expert models. We prove that both linear and non-linear prompts, along with gating functions, can be optimally estimated, with non-linear prompts offering great"},"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":"2502.03029","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-02-05T09:31:27Z","cross_cats_sorted":[],"title_canon_sha256":"f24f663a01ced2bf48f6caa8e14a9b7c582c9ecf8317b1d9e34d4858447e9794","abstract_canon_sha256":"14de0cd5b16221fecbf79843c8a5464aeacd17ea7bcc8408ee094eb1e8274801"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:23:23.095611Z","signature_b64":"IC7EijnmckDBqMG2rFkKQ2w0sxSDd/ZCyuJqKTnwPlToX8UVdEnvGuHIzX0vtQlondCm8EplFOsbHG+YTETJCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"847352098f939406a528d2e26f817b0893d9e58feba2979edc7efa1e18049d9f","last_reissued_at":"2026-07-05T11:23:23.095082Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:23:23.095082Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On Zero-Initialized Attention: Optimal Prompt and Gating Factor Estimation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chau Nguyen, Daniel Sonntag, Duy M. H. Nguyen, Huy Nguyen, Mathias Niepert, Minh Le, Nghiem T. Diep, Nhat Ho","submitted_at":"2025-02-05T09:31:27Z","abstract_excerpt":"The LLaMA-Adapter has recently emerged as an efficient fine-tuning technique for LLaMA models, leveraging zero-initialized attention to stabilize training and enhance performance. However, despite its empirical success, the theoretical foundations of zero-initialized attention remain largely unexplored. In this paper, we provide a rigorous theoretical analysis, establishing a connection between zero-initialized attention and mixture-of-expert models. We prove that both linear and non-linear prompts, along with gating functions, can be optimally estimated, with non-linear prompts offering great"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.03029","kind":"arxiv","version":3},"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/2502.03029/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":"2502.03029","created_at":"2026-07-05T11:23:23.095141+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.03029v3","created_at":"2026-07-05T11:23:23.095141+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.03029","created_at":"2026-07-05T11:23:23.095141+00:00"},{"alias_kind":"pith_short_12","alias_value":"QRZVECMPSOKA","created_at":"2026-07-05T11:23:23.095141+00:00"},{"alias_kind":"pith_short_16","alias_value":"QRZVECMPSOKANJJI","created_at":"2026-07-05T11:23:23.095141+00:00"},{"alias_kind":"pith_short_8","alias_value":"QRZVECMP","created_at":"2026-07-05T11:23:23.095141+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.06845","citing_title":"Convergence Rates for Latent Mixing Measures in Infinite Homoscedastic Location-Scale Mixture Models","ref_index":96,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC","json":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC.json","graph_json":"https://pith.science/api/pith-number/QRZVECMPSOKANJJI2LRG7AL3BC/graph.json","events_json":"https://pith.science/api/pith-number/QRZVECMPSOKANJJI2LRG7AL3BC/events.json","paper":"https://pith.science/paper/QRZVECMP"},"agent_actions":{"view_html":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC","download_json":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC.json","view_paper":"https://pith.science/paper/QRZVECMP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.03029&json=true","fetch_graph":"https://pith.science/api/pith-number/QRZVECMPSOKANJJI2LRG7AL3BC/graph.json","fetch_events":"https://pith.science/api/pith-number/QRZVECMPSOKANJJI2LRG7AL3BC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC/action/storage_attestation","attest_author":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC/action/author_attestation","sign_citation":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC/action/citation_signature","submit_replication":"https://pith.science/pith/QRZVECMPSOKANJJI2LRG7AL3BC/action/replication_record"}},"created_at":"2026-07-05T11:23:23.095141+00:00","updated_at":"2026-07-05T11:23:23.095141+00:00"}