{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:UGQCRQDO4DFH7EPO5MFJGREAZ6","short_pith_number":"pith:UGQCRQDO","schema_version":"1.0","canonical_sha256":"a1a028c06ee0ca7f91eeeb0a934480cf9cc360b0d3718dc6a62deeee3789cff8","source":{"kind":"arxiv","id":"2411.12364","version":2},"attestation_state":"computed","paper":{"title":"Ultra-Sparse Memory Network","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Defa Zhu, Hongzhi Huang, Qiyang Min, Ran Guo, Xun Zhou, Yutao Zeng, Zihao Huang","submitted_at":"2024-11-19T09:24:34Z","abstract_excerpt":"It is widely acknowledged that the performance of Transformer models is logarithmically related to their number of parameters and computational complexity. While approaches like Mixture of Experts (MoE) decouple parameter count from computational complexity, they still face challenges in inference due to high memory access costs. This work introduces UltraMem, incorporating large-scale, ultra-sparse memory layer to address these limitations. Our approach significantly reduces inference latency while maintaining model performance. We also investigate the scaling laws of this new architecture, d"},"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":"2411.12364","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-11-19T09:24:34Z","cross_cats_sorted":[],"title_canon_sha256":"08e4ec7328bb2ac09c6883062512b5ce79410f91ba7d8346c0b9d7e8eeac07b7","abstract_canon_sha256":"5b5b8773ec2c76725702e011b51a006c18b18f6d903bbacc8314b1fca62d659b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:10:29.080321Z","signature_b64":"RTH+yIaAD3OgmCBw79aSteUNfo/o0qfc+6/U1dPNue6QCVKGOPVIbdpdMAR4Mvnzk4rm2BtRQM6vLoqIgZBZDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a1a028c06ee0ca7f91eeeb0a934480cf9cc360b0d3718dc6a62deeee3789cff8","last_reissued_at":"2026-07-05T10:10:29.079853Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:10:29.079853Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Ultra-Sparse Memory Network","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Defa Zhu, Hongzhi Huang, Qiyang Min, Ran Guo, Xun Zhou, Yutao Zeng, Zihao Huang","submitted_at":"2024-11-19T09:24:34Z","abstract_excerpt":"It is widely acknowledged that the performance of Transformer models is logarithmically related to their number of parameters and computational complexity. While approaches like Mixture of Experts (MoE) decouple parameter count from computational complexity, they still face challenges in inference due to high memory access costs. This work introduces UltraMem, incorporating large-scale, ultra-sparse memory layer to address these limitations. Our approach significantly reduces inference latency while maintaining model performance. We also investigate the scaling laws of this new architecture, d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.12364","kind":"arxiv","version":2},"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/2411.12364/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":"2411.12364","created_at":"2026-07-05T10:10:29.079910+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.12364v2","created_at":"2026-07-05T10:10:29.079910+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.12364","created_at":"2026-07-05T10:10:29.079910+00:00"},{"alias_kind":"pith_short_12","alias_value":"UGQCRQDO4DFH","created_at":"2026-07-05T10:10:29.079910+00:00"},{"alias_kind":"pith_short_16","alias_value":"UGQCRQDO4DFH7EPO","created_at":"2026-07-05T10:10:29.079910+00:00"},{"alias_kind":"pith_short_8","alias_value":"UGQCRQDO","created_at":"2026-07-05T10:10:29.079910+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.19172","citing_title":"User as Engram: Internalizing Per-User Memory as Local Parametric Edits","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12113","citing_title":"Augmenting Molecular Language Models with Local $n$-gram Memory","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09888","citing_title":"SinkRec: Mitigating Semantic State Sink in Long Sequence Recommendation with Memory-Conditioned Gated Delta Networks","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20948","citing_title":"Memory Grafting: Scaling Language Model Pre-training via Offline Conditional Memory","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2512.13751","citing_title":"MIDUS: Memory-Infused Depth Up-Scaling","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6","json":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6.json","graph_json":"https://pith.science/api/pith-number/UGQCRQDO4DFH7EPO5MFJGREAZ6/graph.json","events_json":"https://pith.science/api/pith-number/UGQCRQDO4DFH7EPO5MFJGREAZ6/events.json","paper":"https://pith.science/paper/UGQCRQDO"},"agent_actions":{"view_html":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6","download_json":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6.json","view_paper":"https://pith.science/paper/UGQCRQDO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.12364&json=true","fetch_graph":"https://pith.science/api/pith-number/UGQCRQDO4DFH7EPO5MFJGREAZ6/graph.json","fetch_events":"https://pith.science/api/pith-number/UGQCRQDO4DFH7EPO5MFJGREAZ6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6/action/storage_attestation","attest_author":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6/action/author_attestation","sign_citation":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6/action/citation_signature","submit_replication":"https://pith.science/pith/UGQCRQDO4DFH7EPO5MFJGREAZ6/action/replication_record"}},"created_at":"2026-07-05T10:10:29.079910+00:00","updated_at":"2026-07-05T10:10:29.079910+00:00"}