{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:EHDYZX3L2IJWOCDP4MNSVRXDHC","short_pith_number":"pith:EHDYZX3L","schema_version":"1.0","canonical_sha256":"21c78cdf6bd21367086fe31b2ac6e338b14b278af32b8a05d52f7e84ca3b6723","source":{"kind":"arxiv","id":"2606.18811","version":1},"attestation_state":"computed","paper":{"title":"Rescaling MLM-Head for Neural Sparse Retrieval","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.IR","authors_text":"Heuiseok Lim, Jonah Turner, Seongtae Hong, Youngjoon Jang","submitted_at":"2026-06-17T08:27:21Z","abstract_excerpt":"Learned sparse retrieval (LSR) models such as SPLADE have traditionally used BERT-style masked language models as backbone encoders. A natural expectation is that replacing BERT with stronger pretrained encoders should improve retrieval effectiveness. However, we find that under standard SPLADE training recipes, backbones with large MLM-head L2 norms can suffer performance degradation and even training collapse under standard SPLADE training recipes. We identify this failure as a scale mismatch in the MLM head: SPLADE directly uses MLM-head outputs to construct sparse lexical representations, "},"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":"2606.18811","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.IR","submitted_at":"2026-06-17T08:27:21Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c50cd483bc5e18983571828c1d41ad45b6b15cc466ac01121ff4c6122c2a5711","abstract_canon_sha256":"f4fe71f307edaa92b1b739d08d3e7c2d48146803ed3198cb992a75fd16280ce4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:11:48.200070Z","signature_b64":"WbNuTik+K48WP2OZcupG6dMw+HWRs06irNSrOHk5MPMWawLzigleGWlXV6r9oU8GoR60PUG5G0coTLQlAwnCAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"21c78cdf6bd21367086fe31b2ac6e338b14b278af32b8a05d52f7e84ca3b6723","last_reissued_at":"2026-06-19T16:11:48.199709Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:11:48.199709Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rescaling MLM-Head for Neural Sparse Retrieval","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.IR","authors_text":"Heuiseok Lim, Jonah Turner, Seongtae Hong, Youngjoon Jang","submitted_at":"2026-06-17T08:27:21Z","abstract_excerpt":"Learned sparse retrieval (LSR) models such as SPLADE have traditionally used BERT-style masked language models as backbone encoders. A natural expectation is that replacing BERT with stronger pretrained encoders should improve retrieval effectiveness. However, we find that under standard SPLADE training recipes, backbones with large MLM-head L2 norms can suffer performance degradation and even training collapse under standard SPLADE training recipes. We identify this failure as a scale mismatch in the MLM head: SPLADE directly uses MLM-head outputs to construct sparse lexical representations, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.18811","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/2606.18811/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":"2606.18811","created_at":"2026-06-19T16:11:48.199773+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.18811v1","created_at":"2026-06-19T16:11:48.199773+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.18811","created_at":"2026-06-19T16:11:48.199773+00:00"},{"alias_kind":"pith_short_12","alias_value":"EHDYZX3L2IJW","created_at":"2026-06-19T16:11:48.199773+00:00"},{"alias_kind":"pith_short_16","alias_value":"EHDYZX3L2IJWOCDP","created_at":"2026-06-19T16:11:48.199773+00:00"},{"alias_kind":"pith_short_8","alias_value":"EHDYZX3L","created_at":"2026-06-19T16:11:48.199773+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC","json":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC.json","graph_json":"https://pith.science/api/pith-number/EHDYZX3L2IJWOCDP4MNSVRXDHC/graph.json","events_json":"https://pith.science/api/pith-number/EHDYZX3L2IJWOCDP4MNSVRXDHC/events.json","paper":"https://pith.science/paper/EHDYZX3L"},"agent_actions":{"view_html":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC","download_json":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC.json","view_paper":"https://pith.science/paper/EHDYZX3L","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.18811&json=true","fetch_graph":"https://pith.science/api/pith-number/EHDYZX3L2IJWOCDP4MNSVRXDHC/graph.json","fetch_events":"https://pith.science/api/pith-number/EHDYZX3L2IJWOCDP4MNSVRXDHC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC/action/storage_attestation","attest_author":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC/action/author_attestation","sign_citation":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC/action/citation_signature","submit_replication":"https://pith.science/pith/EHDYZX3L2IJWOCDP4MNSVRXDHC/action/replication_record"}},"created_at":"2026-06-19T16:11:48.199773+00:00","updated_at":"2026-06-19T16:11:48.199773+00:00"}