{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:2TJ74S6DGPZA6E4SHG7UFWCI2X","short_pith_number":"pith:2TJ74S6D","schema_version":"1.0","canonical_sha256":"d4d3fe4bc333f20f139239bf42d848d5ef9ae3439f6daec867a44debec39907a","source":{"kind":"arxiv","id":"2305.17212","version":4},"attestation_state":"computed","paper":{"title":"Rotational Equilibrium: How Weight Decay Balances Learning Across Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Atli Kosson, Bettina Messmer, Martin Jaggi","submitted_at":"2023-05-26T19:14:01Z","abstract_excerpt":"This study investigates how weight decay affects the update behavior of individual neurons in deep neural networks through a combination of applied analysis and experimentation. Weight decay can cause the expected magnitude and angular updates of a neuron's weight vector to converge to a steady state we call rotational equilibrium. These states can be highly homogeneous, effectively balancing the average rotation -- a proxy for the effective learning rate -- across different layers and neurons. Our work analyzes these dynamics across optimizers like Adam, Lion, and SGD with momentum, offering "},"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":"2305.17212","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-05-26T19:14:01Z","cross_cats_sorted":[],"title_canon_sha256":"3c4e51142dae7f845c09c14068799757f5bd2aec495d650862447c872391fca4","abstract_canon_sha256":"56821b129875e0eb3e5bbb6c1205dbe3f94146006061f3b65711dca82562dbbe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:26:11.690598Z","signature_b64":"Txo23wK0XWqUeIfVjLhz6jxygljs50/R6ydS7vqonnJpvtfK/fisxTDtvlGA4NS9C+pXUU55Z6/jbt59yZA1Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d4d3fe4bc333f20f139239bf42d848d5ef9ae3439f6daec867a44debec39907a","last_reissued_at":"2026-07-05T08:26:11.690111Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:26:11.690111Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rotational Equilibrium: How Weight Decay Balances Learning Across Neural Networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Atli Kosson, Bettina Messmer, Martin Jaggi","submitted_at":"2023-05-26T19:14:01Z","abstract_excerpt":"This study investigates how weight decay affects the update behavior of individual neurons in deep neural networks through a combination of applied analysis and experimentation. Weight decay can cause the expected magnitude and angular updates of a neuron's weight vector to converge to a steady state we call rotational equilibrium. These states can be highly homogeneous, effectively balancing the average rotation -- a proxy for the effective learning rate -- across different layers and neurons. Our work analyzes these dynamics across optimizers like Adam, Lion, and SGD with momentum, offering "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.17212","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/2305.17212/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":"2305.17212","created_at":"2026-07-05T08:26:11.690169+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.17212v4","created_at":"2026-07-05T08:26:11.690169+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.17212","created_at":"2026-07-05T08:26:11.690169+00:00"},{"alias_kind":"pith_short_12","alias_value":"2TJ74S6DGPZA","created_at":"2026-07-05T08:26:11.690169+00:00"},{"alias_kind":"pith_short_16","alias_value":"2TJ74S6DGPZA6E4S","created_at":"2026-07-05T08:26:11.690169+00:00"},{"alias_kind":"pith_short_8","alias_value":"2TJ74S6D","created_at":"2026-07-05T08:26:11.690169+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25971","citing_title":"Improving Neural Network Training by Decoupling the Magnitude and Direction of Weight Vectors","ref_index":95,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19367","citing_title":"Weibull Weight-Scale Parameter Evolution under AdamW Training Dynamics","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29176","citing_title":"Dead-Direction Conditioners: Gauge-Equivariant Preconditioning for Deep Networks","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16622","citing_title":"Does Weight Decay Enhance Training Stability?","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04418","citing_title":"Demystifying Manifold Constraints in LLM Pre-training","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22838","citing_title":"Neural Network Optimization Reimagined: Decoupled Techniques for Scratch and Fine-Tuning","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X","json":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X.json","graph_json":"https://pith.science/api/pith-number/2TJ74S6DGPZA6E4SHG7UFWCI2X/graph.json","events_json":"https://pith.science/api/pith-number/2TJ74S6DGPZA6E4SHG7UFWCI2X/events.json","paper":"https://pith.science/paper/2TJ74S6D"},"agent_actions":{"view_html":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X","download_json":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X.json","view_paper":"https://pith.science/paper/2TJ74S6D","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.17212&json=true","fetch_graph":"https://pith.science/api/pith-number/2TJ74S6DGPZA6E4SHG7UFWCI2X/graph.json","fetch_events":"https://pith.science/api/pith-number/2TJ74S6DGPZA6E4SHG7UFWCI2X/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X/action/storage_attestation","attest_author":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X/action/author_attestation","sign_citation":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X/action/citation_signature","submit_replication":"https://pith.science/pith/2TJ74S6DGPZA6E4SHG7UFWCI2X/action/replication_record"}},"created_at":"2026-07-05T08:26:11.690169+00:00","updated_at":"2026-07-05T08:26:11.690169+00:00"}