{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:342CXSHFMIORW5DWN2GOXQDZQD","short_pith_number":"pith:342CXSHF","schema_version":"1.0","canonical_sha256":"df342bc8e5621d1b74766e8cebc07980c086d38d32a79624f887db29a6b702aa","source":{"kind":"arxiv","id":"2505.24688","version":4},"attestation_state":"computed","paper":{"title":"Soft Reasoning: Navigating Solution Spaces in Large Language Models through Controlled Embedding Exploration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hanqi Yan, Lin Gui, Qinglin Zhu, Runcong Zhao, Yudong Chen, Yulan He","submitted_at":"2025-05-30T15:11:52Z","abstract_excerpt":"Large Language Models (LLMs) struggle with complex reasoning due to limited diversity and inefficient search. We propose Soft Reasoning, an embedding-based search framework that optimises the embedding of the first token to guide generation. It combines (1) embedding perturbation for controlled exploration and (2) Bayesian optimisation to refine embeddings via a verifier-guided objective, balancing exploration and exploitation. This approach improves reasoning accuracy and coherence while avoiding reliance on heuristic search. Experiments demonstrate superior correctness with minimal computati"},"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":"2505.24688","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-30T15:11:52Z","cross_cats_sorted":[],"title_canon_sha256":"26b77e0ba2d202a1a9a33d50c6c8f80d6d045c7dbab971fd1a89702d397831fa","abstract_canon_sha256":"02143bee8a62b86a9c81a38d9f0690bc20201480c05bc48ee6d16890e5daa521"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:11:18.343450Z","signature_b64":"yPlJidBgLP7mq5kT6qtH2NxhNagd88xJfNt2Y6UL4ljE+hfg17UOWjbIOP5k4+7DAfVuoBvd5cFrWjhBMN/aDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"df342bc8e5621d1b74766e8cebc07980c086d38d32a79624f887db29a6b702aa","last_reissued_at":"2026-07-05T12:11:18.342875Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:11:18.342875Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Soft Reasoning: Navigating Solution Spaces in Large Language Models through Controlled Embedding Exploration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hanqi Yan, Lin Gui, Qinglin Zhu, Runcong Zhao, Yudong Chen, Yulan He","submitted_at":"2025-05-30T15:11:52Z","abstract_excerpt":"Large Language Models (LLMs) struggle with complex reasoning due to limited diversity and inefficient search. We propose Soft Reasoning, an embedding-based search framework that optimises the embedding of the first token to guide generation. It combines (1) embedding perturbation for controlled exploration and (2) Bayesian optimisation to refine embeddings via a verifier-guided objective, balancing exploration and exploitation. This approach improves reasoning accuracy and coherence while avoiding reliance on heuristic search. Experiments demonstrate superior correctness with minimal computati"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.24688","kind":"arxiv","version":4},"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/2505.24688/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":"2505.24688","created_at":"2026-07-05T12:11:18.342940+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.24688v4","created_at":"2026-07-05T12:11:18.342940+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.24688","created_at":"2026-07-05T12:11:18.342940+00:00"},{"alias_kind":"pith_short_12","alias_value":"342CXSHFMIOR","created_at":"2026-07-05T12:11:18.342940+00:00"},{"alias_kind":"pith_short_16","alias_value":"342CXSHFMIORW5DW","created_at":"2026-07-05T12:11:18.342940+00:00"},{"alias_kind":"pith_short_8","alias_value":"342CXSHF","created_at":"2026-07-05T12:11:18.342940+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10768","citing_title":"N-GRPO: Embedding-Level Neighbor Mixing for Enhanced Policy Optimization","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD","json":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD.json","graph_json":"https://pith.science/api/pith-number/342CXSHFMIORW5DWN2GOXQDZQD/graph.json","events_json":"https://pith.science/api/pith-number/342CXSHFMIORW5DWN2GOXQDZQD/events.json","paper":"https://pith.science/paper/342CXSHF"},"agent_actions":{"view_html":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD","download_json":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD.json","view_paper":"https://pith.science/paper/342CXSHF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.24688&json=true","fetch_graph":"https://pith.science/api/pith-number/342CXSHFMIORW5DWN2GOXQDZQD/graph.json","fetch_events":"https://pith.science/api/pith-number/342CXSHFMIORW5DWN2GOXQDZQD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD/action/storage_attestation","attest_author":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD/action/author_attestation","sign_citation":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD/action/citation_signature","submit_replication":"https://pith.science/pith/342CXSHFMIORW5DWN2GOXQDZQD/action/replication_record"}},"created_at":"2026-07-05T12:11:18.342940+00:00","updated_at":"2026-07-05T12:11:18.342940+00:00"}