{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:W7D3JMFUX5SNEQGYLGSRFFZHJ5","short_pith_number":"pith:W7D3JMFU","schema_version":"1.0","canonical_sha256":"b7c7b4b0b4bf64d240d859a51297274f64fab788368c36632b62b43d1ab1030c","source":{"kind":"arxiv","id":"2306.03745","version":2},"attestation_state":"computed","paper":{"title":"Soft Merging of Experts with Adaptive Routing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Colin Raffel, Haokun Liu, Mohammed Muqeeth","submitted_at":"2023-06-06T15:04:31Z","abstract_excerpt":"Sparsely activated neural networks with conditional computation learn to route their inputs through different \"expert\" subnetworks, providing a form of modularity that densely activated models lack. Despite their possible benefits, models with learned routing often underperform their parameter-matched densely activated counterparts as well as models that use non-learned heuristic routing strategies. In this paper, we hypothesize that these shortcomings stem from the gradient estimation techniques used to train sparsely activated models that use non-differentiable discrete routing decisions. To"},"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":"2306.03745","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-06-06T15:04:31Z","cross_cats_sorted":[],"title_canon_sha256":"048043caaa1893c7dd57a34e86a1e6eff88ea97aeb5a239bdc811764c11d3272","abstract_canon_sha256":"0f90d2cd8f24adf5b5eeda316e9154764d3efcea56ec59f0219de210086d340c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:18:31.970774Z","signature_b64":"72GfQGbzbEptCz+IphGDbz/BSfpWjIDATVhy+BidfGyH6qyYKFRRzjCUIvjnNtESPsnGAK9fRe7wMdxmWBM4CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b7c7b4b0b4bf64d240d859a51297274f64fab788368c36632b62b43d1ab1030c","last_reissued_at":"2026-07-05T08:18:31.970359Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:18:31.970359Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Soft Merging of Experts with Adaptive Routing","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Colin Raffel, Haokun Liu, Mohammed Muqeeth","submitted_at":"2023-06-06T15:04:31Z","abstract_excerpt":"Sparsely activated neural networks with conditional computation learn to route their inputs through different \"expert\" subnetworks, providing a form of modularity that densely activated models lack. Despite their possible benefits, models with learned routing often underperform their parameter-matched densely activated counterparts as well as models that use non-learned heuristic routing strategies. In this paper, we hypothesize that these shortcomings stem from the gradient estimation techniques used to train sparsely activated models that use non-differentiable discrete routing decisions. To"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.03745","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/2306.03745/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":"2306.03745","created_at":"2026-07-05T08:18:31.970422+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.03745v2","created_at":"2026-07-05T08:18:31.970422+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.03745","created_at":"2026-07-05T08:18:31.970422+00:00"},{"alias_kind":"pith_short_12","alias_value":"W7D3JMFUX5SN","created_at":"2026-07-05T08:18:31.970422+00:00"},{"alias_kind":"pith_short_16","alias_value":"W7D3JMFUX5SNEQGY","created_at":"2026-07-05T08:18:31.970422+00:00"},{"alias_kind":"pith_short_8","alias_value":"W7D3JMFU","created_at":"2026-07-05T08:18:31.970422+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":154,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":154,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10933","citing_title":"DECO: Sparse Mixture-of-Experts with Dense-Comparable Performance on End-Side Devices","ref_index":154,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21330","citing_title":"Teacher-Guided Routing for Sparse Vision Mixture-of-Experts","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5","json":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5.json","graph_json":"https://pith.science/api/pith-number/W7D3JMFUX5SNEQGYLGSRFFZHJ5/graph.json","events_json":"https://pith.science/api/pith-number/W7D3JMFUX5SNEQGYLGSRFFZHJ5/events.json","paper":"https://pith.science/paper/W7D3JMFU"},"agent_actions":{"view_html":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5","download_json":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5.json","view_paper":"https://pith.science/paper/W7D3JMFU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.03745&json=true","fetch_graph":"https://pith.science/api/pith-number/W7D3JMFUX5SNEQGYLGSRFFZHJ5/graph.json","fetch_events":"https://pith.science/api/pith-number/W7D3JMFUX5SNEQGYLGSRFFZHJ5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5/action/storage_attestation","attest_author":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5/action/author_attestation","sign_citation":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5/action/citation_signature","submit_replication":"https://pith.science/pith/W7D3JMFUX5SNEQGYLGSRFFZHJ5/action/replication_record"}},"created_at":"2026-07-05T08:18:31.970422+00:00","updated_at":"2026-07-05T08:18:31.970422+00:00"}