{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:5Q2EVXA4BWUQQST7DYBLQUSOUA","short_pith_number":"pith:5Q2EVXA4","schema_version":"1.0","canonical_sha256":"ec344adc1c0da9084a7f1e02b8524ea00fe7ea2e92d8a0d7bcc4a83177af890b","source":{"kind":"arxiv","id":"2606.19491","version":1},"attestation_state":"computed","paper":{"title":"Algebraic Dead Directions in LayerNorm Transformers: A Forward-Pass-Only Diagnostic at LLM Scale","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"P. J. Narayanan, Tejas Pradeep Shirodkar","submitted_at":"2026-06-17T18:28:37Z","abstract_excerpt":"Pretrained transformers sit near singular minima of the loss, where the Fisher information metric degenerates along dead directions: directions in parameter space along which the directional Fisher vanishes. Locating such a direction normally needs a forward pass and an eigendecomposition of activations, or a sampling-based complexity estimate; none returns a direction computable from the network's parameters alone. We give one, for LayerNorm transformers. The inverse-scale direction $\\gamma^{-1}/\\|\\gamma^{-1}\\|$ of the LayerNorm affine is an exact algebraic kernel of the post-final-norm centr"},"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":"2606.19491","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-17T18:28:37Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"b746e0bbe2639cdf1da4bd16ec61e9073259dd41ee0325c06cadc07baf569a9a","abstract_canon_sha256":"902c2601f126159ad7aae3c0a7244811506ee7f9328b5a576a366199db9b0713"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:12:27.171077Z","signature_b64":"Aobktxy9BynKUjdzUNVkPClazqkLGSkZQUospT2+eYgCpYI2tZ5KtcIs//YClLd+F7JrhpTbo1vKHeSzDH1AAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ec344adc1c0da9084a7f1e02b8524ea00fe7ea2e92d8a0d7bcc4a83177af890b","last_reissued_at":"2026-06-19T16:12:27.170737Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:12:27.170737Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Algebraic Dead Directions in LayerNorm Transformers: A Forward-Pass-Only Diagnostic at LLM Scale","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"P. J. Narayanan, Tejas Pradeep Shirodkar","submitted_at":"2026-06-17T18:28:37Z","abstract_excerpt":"Pretrained transformers sit near singular minima of the loss, where the Fisher information metric degenerates along dead directions: directions in parameter space along which the directional Fisher vanishes. Locating such a direction normally needs a forward pass and an eigendecomposition of activations, or a sampling-based complexity estimate; none returns a direction computable from the network's parameters alone. We give one, for LayerNorm transformers. The inverse-scale direction $\\gamma^{-1}/\\|\\gamma^{-1}\\|$ of the LayerNorm affine is an exact algebraic kernel of the post-final-norm centr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.19491","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/2606.19491/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":"2606.19491","created_at":"2026-06-19T16:12:27.170798+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.19491v1","created_at":"2026-06-19T16:12:27.170798+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.19491","created_at":"2026-06-19T16:12:27.170798+00:00"},{"alias_kind":"pith_short_12","alias_value":"5Q2EVXA4BWUQ","created_at":"2026-06-19T16:12:27.170798+00:00"},{"alias_kind":"pith_short_16","alias_value":"5Q2EVXA4BWUQQST7","created_at":"2026-06-19T16:12:27.170798+00:00"},{"alias_kind":"pith_short_8","alias_value":"5Q2EVXA4","created_at":"2026-06-19T16:12:27.170798+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2606.29176","citing_title":"Dead-Direction Conditioners: Gauge-Equivariant Preconditioning for Deep Networks","ref_index":30,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA","json":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA.json","graph_json":"https://pith.science/api/pith-number/5Q2EVXA4BWUQQST7DYBLQUSOUA/graph.json","events_json":"https://pith.science/api/pith-number/5Q2EVXA4BWUQQST7DYBLQUSOUA/events.json","paper":"https://pith.science/paper/5Q2EVXA4"},"agent_actions":{"view_html":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA","download_json":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA.json","view_paper":"https://pith.science/paper/5Q2EVXA4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.19491&json=true","fetch_graph":"https://pith.science/api/pith-number/5Q2EVXA4BWUQQST7DYBLQUSOUA/graph.json","fetch_events":"https://pith.science/api/pith-number/5Q2EVXA4BWUQQST7DYBLQUSOUA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA/action/storage_attestation","attest_author":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA/action/author_attestation","sign_citation":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA/action/citation_signature","submit_replication":"https://pith.science/pith/5Q2EVXA4BWUQQST7DYBLQUSOUA/action/replication_record"}},"created_at":"2026-06-19T16:12:27.170798+00:00","updated_at":"2026-06-19T16:12:27.170798+00:00"}