{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:JPKXLFDROQIYCBQG6LWOMOOJAE","short_pith_number":"pith:JPKXLFDR","schema_version":"1.0","canonical_sha256":"4bd57594717411810606f2ece639c901327d1bb998241fb2a4d79fec74810dd1","source":{"kind":"arxiv","id":"2310.10837","version":3},"attestation_state":"computed","paper":{"title":"Approximating Two-Layer Feedforward Networks for Efficient Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NE"],"primary_cat":"cs.LG","authors_text":"J\\\"urgen Schmidhuber, Kazuki Irie, R\\'obert Csord\\'as","submitted_at":"2023-10-16T21:23:16Z","abstract_excerpt":"How to reduce compute and memory requirements of neural networks (NNs) without sacrificing performance? Many recent works use sparse Mixtures of Experts (MoEs) to build resource-efficient large language models (LMs). Here we introduce several novel perspectives on MoEs, presenting a general framework that unifies various methods to approximate two-layer NNs (e.g., feedforward blocks of Transformers), including product-key memories (PKMs). Leveraging insights from this framework, we propose methods to improve both MoEs and PKMs. Unlike prior work that compares MoEs with dense baselines under th"},"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":"2310.10837","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-16T21:23:16Z","cross_cats_sorted":["cs.NE"],"title_canon_sha256":"dd28295ef045e16e512b74e472d8fbd4bb9329e383c3c544bf793e052343985b","abstract_canon_sha256":"23a6ba15120ebe570174ec2dc648fed9fe873ce1c04a1fcf7ab34d57f85254c1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:14:57.388869Z","signature_b64":"U1gT3Ygm+ufWRCtoZt8UeTrYZq3XRgEHedBz1sJRVlQyboO4ochoSmUjSVfKJQKUQMoexomE5RMMhphBjjj3CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4bd57594717411810606f2ece639c901327d1bb998241fb2a4d79fec74810dd1","last_reissued_at":"2026-07-05T07:14:57.388450Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:14:57.388450Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Approximating Two-Layer Feedforward Networks for Efficient Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NE"],"primary_cat":"cs.LG","authors_text":"J\\\"urgen Schmidhuber, Kazuki Irie, R\\'obert Csord\\'as","submitted_at":"2023-10-16T21:23:16Z","abstract_excerpt":"How to reduce compute and memory requirements of neural networks (NNs) without sacrificing performance? Many recent works use sparse Mixtures of Experts (MoEs) to build resource-efficient large language models (LMs). Here we introduce several novel perspectives on MoEs, presenting a general framework that unifies various methods to approximate two-layer NNs (e.g., feedforward blocks of Transformers), including product-key memories (PKMs). Leveraging insights from this framework, we propose methods to improve both MoEs and PKMs. Unlike prior work that compares MoEs with dense baselines under th"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.10837","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/2310.10837/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":"2310.10837","created_at":"2026-07-05T07:14:57.388510+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.10837v3","created_at":"2026-07-05T07:14:57.388510+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.10837","created_at":"2026-07-05T07:14:57.388510+00:00"},{"alias_kind":"pith_short_12","alias_value":"JPKXLFDROQIY","created_at":"2026-07-05T07:14:57.388510+00:00"},{"alias_kind":"pith_short_16","alias_value":"JPKXLFDROQIYCBQG","created_at":"2026-07-05T07:14:57.388510+00:00"},{"alias_kind":"pith_short_8","alias_value":"JPKXLFDR","created_at":"2026-07-05T07:14:57.388510+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09165","citing_title":"Sparse Layers are Critical to Scaling Looped Language Models","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2505.06708","citing_title":"Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09165","citing_title":"Sparse Layers are Critical to Scaling Looped Language Models","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE","json":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE.json","graph_json":"https://pith.science/api/pith-number/JPKXLFDROQIYCBQG6LWOMOOJAE/graph.json","events_json":"https://pith.science/api/pith-number/JPKXLFDROQIYCBQG6LWOMOOJAE/events.json","paper":"https://pith.science/paper/JPKXLFDR"},"agent_actions":{"view_html":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE","download_json":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE.json","view_paper":"https://pith.science/paper/JPKXLFDR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.10837&json=true","fetch_graph":"https://pith.science/api/pith-number/JPKXLFDROQIYCBQG6LWOMOOJAE/graph.json","fetch_events":"https://pith.science/api/pith-number/JPKXLFDROQIYCBQG6LWOMOOJAE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE/action/storage_attestation","attest_author":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE/action/author_attestation","sign_citation":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE/action/citation_signature","submit_replication":"https://pith.science/pith/JPKXLFDROQIYCBQG6LWOMOOJAE/action/replication_record"}},"created_at":"2026-07-05T07:14:57.388510+00:00","updated_at":"2026-07-05T07:14:57.388510+00:00"}