{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DZOW5YAZHPVELONQ6HGCJ6LVUV","short_pith_number":"pith:DZOW5YAZ","schema_version":"1.0","canonical_sha256":"1e5d6ee0193bea45b9b0f1cc24f975a546be38bbbc64b3a2e22dd54c29688b91","source":{"kind":"arxiv","id":"2505.19475","version":2},"attestation_state":"computed","paper":{"title":"Continuous Self-Improvement of Large Language Models by Test-time Training with Verifier-Driven Sample Selection","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hossam Amer, Mohammad Mahdi Moradi, Sudhir Mudur, Walid Ahmed, Weiwei Zhang, Yang Liu","submitted_at":"2025-05-26T03:54:47Z","abstract_excerpt":"Learning to adapt pretrained language models to unlabeled, out-of-distribution data is a critical challenge, as models often falter on structurally novel reasoning tasks even while excelling within their training distribution. We introduce a new framework called VDS-TTT - Verifier-Driven Sample Selection for Test-Time Training to efficiently address this. We use a learned verifier to score a pool of generated responses and select only from high ranking pseudo-labeled examples for fine-tuned adaptation. Specifically, for each input query our LLM generates N candidate answers; the verifier assig"},"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":"2505.19475","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-26T03:54:47Z","cross_cats_sorted":[],"title_canon_sha256":"eba24d625903d134b280e1415f400fc13eed36f160d4db8a903eaf7730f711e1","abstract_canon_sha256":"72325415a31ec0f7e64024e94fea001018a64887864c054b021cc79ba194a2a0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:58.214373Z","signature_b64":"W5NIUBnvhGkHHu/hl/VRuE2Ed/TgsrCIgUGDUdv1CL9yCdXIqgj8843PeuVGekD+syp9VuQocCLY+jFc/ERmCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1e5d6ee0193bea45b9b0f1cc24f975a546be38bbbc64b3a2e22dd54c29688b91","last_reissued_at":"2026-07-05T11:10:58.213949Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:58.213949Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Continuous Self-Improvement of Large Language Models by Test-time Training with Verifier-Driven Sample Selection","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hossam Amer, Mohammad Mahdi Moradi, Sudhir Mudur, Walid Ahmed, Weiwei Zhang, Yang Liu","submitted_at":"2025-05-26T03:54:47Z","abstract_excerpt":"Learning to adapt pretrained language models to unlabeled, out-of-distribution data is a critical challenge, as models often falter on structurally novel reasoning tasks even while excelling within their training distribution. We introduce a new framework called VDS-TTT - Verifier-Driven Sample Selection for Test-Time Training to efficiently address this. We use a learned verifier to score a pool of generated responses and select only from high ranking pseudo-labeled examples for fine-tuned adaptation. Specifically, for each input query our LLM generates N candidate answers; the verifier assig"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.19475","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/2505.19475/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":"2505.19475","created_at":"2026-07-05T11:10:58.213999+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.19475v2","created_at":"2026-07-05T11:10:58.213999+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.19475","created_at":"2026-07-05T11:10:58.213999+00:00"},{"alias_kind":"pith_short_12","alias_value":"DZOW5YAZHPVE","created_at":"2026-07-05T11:10:58.213999+00:00"},{"alias_kind":"pith_short_16","alias_value":"DZOW5YAZHPVELONQ","created_at":"2026-07-05T11:10:58.213999+00:00"},{"alias_kind":"pith_short_8","alias_value":"DZOW5YAZ","created_at":"2026-07-05T11:10:58.213999+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.20189","citing_title":"SOLAR: A Self-Optimizing Open-Ended Autonomous Agent for Lifelong Learning and Continual Adaptation","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11328","citing_title":"Epistemic Uncertainty for Test-Time Discovery","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18131","citing_title":"Training LLM Agents for Spontaneous, Reward-Free Self-Evolution via World Knowledge Exploration","ref_index":43,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV","json":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV.json","graph_json":"https://pith.science/api/pith-number/DZOW5YAZHPVELONQ6HGCJ6LVUV/graph.json","events_json":"https://pith.science/api/pith-number/DZOW5YAZHPVELONQ6HGCJ6LVUV/events.json","paper":"https://pith.science/paper/DZOW5YAZ"},"agent_actions":{"view_html":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV","download_json":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV.json","view_paper":"https://pith.science/paper/DZOW5YAZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.19475&json=true","fetch_graph":"https://pith.science/api/pith-number/DZOW5YAZHPVELONQ6HGCJ6LVUV/graph.json","fetch_events":"https://pith.science/api/pith-number/DZOW5YAZHPVELONQ6HGCJ6LVUV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV/action/storage_attestation","attest_author":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV/action/author_attestation","sign_citation":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV/action/citation_signature","submit_replication":"https://pith.science/pith/DZOW5YAZHPVELONQ6HGCJ6LVUV/action/replication_record"}},"created_at":"2026-07-05T11:10:58.213999+00:00","updated_at":"2026-07-05T11:10:58.213999+00:00"}