{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:3F25PURSCHMF4IRBG52QAHVEHU","short_pith_number":"pith:3F25PURS","schema_version":"1.0","canonical_sha256":"d975d7d23211d85e22213775001ea43d1c49007ca69956a688ce63d93ada91e8","source":{"kind":"arxiv","id":"2503.01843","version":3},"attestation_state":"computed","paper":{"title":"When Can You Get Away with Low Memory Adam?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","stat.ML"],"primary_cat":"cs.LG","authors_text":"Dayal Singh Kalra, John Kirchenbauer, Maissam Barkeshli, Tom Goldstein","submitted_at":"2025-03-03T18:59:40Z","abstract_excerpt":"Adam is the go-to optimizer for training modern machine learning models, but it requires additional memory to maintain the moving averages of the gradients and their squares. While various low-memory optimizers have been proposed that sometimes match the performance of Adam, their lack of reliability has left Adam as the default choice. In this work, we apply a simple layer-wise Signal-to-Noise Ratio (SNR) analysis to quantify when second-moment tensors can be effectively replaced by their means across different dimensions. Our SNR analysis reveals how architecture, training hyperparameters, a"},"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":"2503.01843","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-03T18:59:40Z","cross_cats_sorted":["cond-mat.dis-nn","stat.ML"],"title_canon_sha256":"7a840afe3bf29e490514babdaad1ef23f2b9fc74ad20011d333aa37f651f3429","abstract_canon_sha256":"cbec4de1bf8506633ba9da276e86fe598510e4b53546eebe15109f326641b31b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:33:08.933463Z","signature_b64":"EOwhQdvRVwT5W1k7Zm9ZbHPYR+nzNTsPIlYafucK5kML6U0vK6bWTaYRZFEkSh0nkxeWmSmK9UJAclL7/V5OCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d975d7d23211d85e22213775001ea43d1c49007ca69956a688ce63d93ada91e8","last_reissued_at":"2026-07-05T10:33:08.932392Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:33:08.932392Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Can You Get Away with Low Memory Adam?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","stat.ML"],"primary_cat":"cs.LG","authors_text":"Dayal Singh Kalra, John Kirchenbauer, Maissam Barkeshli, Tom Goldstein","submitted_at":"2025-03-03T18:59:40Z","abstract_excerpt":"Adam is the go-to optimizer for training modern machine learning models, but it requires additional memory to maintain the moving averages of the gradients and their squares. While various low-memory optimizers have been proposed that sometimes match the performance of Adam, their lack of reliability has left Adam as the default choice. In this work, we apply a simple layer-wise Signal-to-Noise Ratio (SNR) analysis to quantify when second-moment tensors can be effectively replaced by their means across different dimensions. Our SNR analysis reveals how architecture, training hyperparameters, a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.01843","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/2503.01843/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":"2503.01843","created_at":"2026-07-05T10:33:08.932529+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.01843v3","created_at":"2026-07-05T10:33:08.932529+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.01843","created_at":"2026-07-05T10:33:08.932529+00:00"},{"alias_kind":"pith_short_12","alias_value":"3F25PURSCHMF","created_at":"2026-07-05T10:33:08.932529+00:00"},{"alias_kind":"pith_short_16","alias_value":"3F25PURSCHMF4IRB","created_at":"2026-07-05T10:33:08.932529+00:00"},{"alias_kind":"pith_short_8","alias_value":"3F25PURS","created_at":"2026-07-05T10:33:08.932529+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2506.16659","citing_title":"Memory-Efficient LLM Pretraining via Minimalist Optimizer Design","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU","json":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU.json","graph_json":"https://pith.science/api/pith-number/3F25PURSCHMF4IRBG52QAHVEHU/graph.json","events_json":"https://pith.science/api/pith-number/3F25PURSCHMF4IRBG52QAHVEHU/events.json","paper":"https://pith.science/paper/3F25PURS"},"agent_actions":{"view_html":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU","download_json":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU.json","view_paper":"https://pith.science/paper/3F25PURS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.01843&json=true","fetch_graph":"https://pith.science/api/pith-number/3F25PURSCHMF4IRBG52QAHVEHU/graph.json","fetch_events":"https://pith.science/api/pith-number/3F25PURSCHMF4IRBG52QAHVEHU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU/action/storage_attestation","attest_author":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU/action/author_attestation","sign_citation":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU/action/citation_signature","submit_replication":"https://pith.science/pith/3F25PURSCHMF4IRBG52QAHVEHU/action/replication_record"}},"created_at":"2026-07-05T10:33:08.932529+00:00","updated_at":"2026-07-05T10:33:08.932529+00:00"}