{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OPJ6GYGWLRUJDPW4GOWJSNPW3J","short_pith_number":"pith:OPJ6GYGW","schema_version":"1.0","canonical_sha256":"73d3e360d65c6891bedc33ac9935f6da5a331d1656f8a8742c5312faa80919a7","source":{"kind":"arxiv","id":"2505.19488","version":1},"attestation_state":"computed","paper":{"title":"Understanding Transformer from the Perspective of Associative Memory","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Guang Shi, Mingyu Xu, Shu Zhong, Tenglong Ao","submitted_at":"2025-05-26T04:15:38Z","abstract_excerpt":"In this paper, we share our reflections and insights on understanding Transformer architectures through the lens of associative memory--a classic psychological concept inspired by human cognition. We start with the basics of associative memory (think simple linear attention) and then dive into two dimensions:\n  Memory Capacity: How much can a Transformer really remember, and how well? We introduce retrieval SNR to measure this and use a kernel perspective to mathematically reveal why Softmax Attention is so effective. We also show how FFNs can be seen as a type of associative memory, leading t"},"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":"2505.19488","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-26T04:15:38Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0956e1d5dbb62cf03be573cdd50f75f45a7d96e17dc18b08bc4ff4461c5261f5","abstract_canon_sha256":"2043d681789825b087874499c504fcbba739868717f1ef0dc4d7faae60294987"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:31.084771Z","signature_b64":"wQOSvZCphlR/TZhpO4V3ZSK4IVGGZ734b5jRxiyr99bR9cfUnoe/BNXm42zREgCdPYJF8dXFM+heZyy+m3cZCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"73d3e360d65c6891bedc33ac9935f6da5a331d1656f8a8742c5312faa80919a7","last_reissued_at":"2026-07-05T11:09:31.084279Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:31.084279Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Transformer from the Perspective of Associative Memory","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Guang Shi, Mingyu Xu, Shu Zhong, Tenglong Ao","submitted_at":"2025-05-26T04:15:38Z","abstract_excerpt":"In this paper, we share our reflections and insights on understanding Transformer architectures through the lens of associative memory--a classic psychological concept inspired by human cognition. We start with the basics of associative memory (think simple linear attention) and then dive into two dimensions:\n  Memory Capacity: How much can a Transformer really remember, and how well? We introduce retrieval SNR to measure this and use a kernel perspective to mathematically reveal why Softmax Attention is so effective. We also show how FFNs can be seen as a type of associative memory, leading t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.19488","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/2505.19488/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":"2505.19488","created_at":"2026-07-05T11:09:31.084338+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.19488v1","created_at":"2026-07-05T11:09:31.084338+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.19488","created_at":"2026-07-05T11:09:31.084338+00:00"},{"alias_kind":"pith_short_12","alias_value":"OPJ6GYGWLRUJ","created_at":"2026-07-05T11:09:31.084338+00:00"},{"alias_kind":"pith_short_16","alias_value":"OPJ6GYGWLRUJDPW4","created_at":"2026-07-05T11:09:31.084338+00:00"},{"alias_kind":"pith_short_8","alias_value":"OPJ6GYGW","created_at":"2026-07-05T11:09:31.084338+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.31830","citing_title":"PriorEye: Geospatial Visual Priors for End-to-End Autonomous Driving","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26647","citing_title":"More Expressive Feedforward Layers: Part I. Token-Adaptive Mixing of Activations","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00620","citing_title":"FlowNar: Scalable Streaming Narration for Long-Form Videos","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06034","citing_title":"When Good Enough Is Optimal: Multiplication-Only Matrix Inversion Approximation for Quantized Gated DeltaNet","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2603.15031","citing_title":"Attention Residuals","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2509.22321","citing_title":"Distributed Associative Memory via Online Convex Optimization","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2509.24552","citing_title":"Short window attention enables long-term memorization","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2602.21204","citing_title":"Test-Time Training with KV Binding Is Secretly Linear Attention","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2510.25741","citing_title":"Scaling Latent Reasoning via Looped Language Models","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2510.26692","citing_title":"Kimi Linear: An Expressive, Efficient Attention Architecture","ref_index":128,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06665","citing_title":"UniPool: A Globally Shared Expert Pool for Mixture-of-Experts","ref_index":63,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05838","citing_title":"MDN: Parallelizing Stepwise Momentum for Delta Linear Attention","ref_index":85,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J","json":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J.json","graph_json":"https://pith.science/api/pith-number/OPJ6GYGWLRUJDPW4GOWJSNPW3J/graph.json","events_json":"https://pith.science/api/pith-number/OPJ6GYGWLRUJDPW4GOWJSNPW3J/events.json","paper":"https://pith.science/paper/OPJ6GYGW"},"agent_actions":{"view_html":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J","download_json":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J.json","view_paper":"https://pith.science/paper/OPJ6GYGW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.19488&json=true","fetch_graph":"https://pith.science/api/pith-number/OPJ6GYGWLRUJDPW4GOWJSNPW3J/graph.json","fetch_events":"https://pith.science/api/pith-number/OPJ6GYGWLRUJDPW4GOWJSNPW3J/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J/action/storage_attestation","attest_author":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J/action/author_attestation","sign_citation":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J/action/citation_signature","submit_replication":"https://pith.science/pith/OPJ6GYGWLRUJDPW4GOWJSNPW3J/action/replication_record"}},"created_at":"2026-07-05T11:09:31.084338+00:00","updated_at":"2026-07-05T11:09:31.084338+00:00"}