{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NPKJA5VN3H4NDID6D744RVNIP4","short_pith_number":"pith:NPKJA5VN","schema_version":"1.0","canonical_sha256":"6bd49076add9f8d1a07e1ff9c8d5a87f0d0fa546e0967f9aed7bd99e06996f7e","source":{"kind":"arxiv","id":"2608.03204","version":1},"attestation_state":"computed","paper":{"title":"Aligning Large Vision-Language Models at Test Time: A Trajectory-Guided Structured Sampling Approach","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gerard de Melo, Linlin Wang, Tianbao Jiang, Weicong Ni","submitted_at":"2026-08-04T06:47:08Z","abstract_excerpt":"Post-training reinforcement learning (RL) algorithms are commonly used to align large vision-language models (LVLMs) with human intent and the requirements of visual reasoning tasks. However, existing RL-based alignment methods are often resource-intensive and encounter mismatches between training objectives and inference-time distributions. To bridge this gap, we propose a novel test-time alignment approach that leverages trajectory-guided structured sampling for dynamic inference-time refinement, achieving better alignment with visual grounding and ensuring logical consistency. Our approach "},"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":"2608.03204","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-08-04T06:47:08Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"108058e5ee814b48eb5428a85666c30f8880ffc9106ae2da2f981c5f5d353286","abstract_canon_sha256":"a15a1370ec5982d255fbea0e0d0378f47ef31357945dd39ab0ebb72111ddfc6c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-05T00:46:09.931805Z","signature_b64":"+pBoHRFegjzWT9KPnAwQLUjOS+Dj63A9ochfkOcVeOqkoFLJdVxXsb70lqJkHO0V/+5Ac+IZq8f+R87/GNC/AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6bd49076add9f8d1a07e1ff9c8d5a87f0d0fa546e0967f9aed7bd99e06996f7e","last_reissued_at":"2026-08-05T00:46:09.929165Z","signature_status":"signed_v1","first_computed_at":"2026-08-05T00:46:09.929165Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Aligning Large Vision-Language Models at Test Time: A Trajectory-Guided Structured Sampling Approach","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Gerard de Melo, Linlin Wang, Tianbao Jiang, Weicong Ni","submitted_at":"2026-08-04T06:47:08Z","abstract_excerpt":"Post-training reinforcement learning (RL) algorithms are commonly used to align large vision-language models (LVLMs) with human intent and the requirements of visual reasoning tasks. However, existing RL-based alignment methods are often resource-intensive and encounter mismatches between training objectives and inference-time distributions. To bridge this gap, we propose a novel test-time alignment approach that leverages trajectory-guided structured sampling for dynamic inference-time refinement, achieving better alignment with visual grounding and ensuring logical consistency. Our approach "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.03204","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/2608.03204/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":"2608.03204","created_at":"2026-08-05T00:46:09.930210+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.03204v1","created_at":"2026-08-05T00:46:09.930210+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.03204","created_at":"2026-08-05T00:46:09.930210+00:00"},{"alias_kind":"pith_short_12","alias_value":"NPKJA5VN3H4N","created_at":"2026-08-05T00:46:09.930210+00:00"},{"alias_kind":"pith_short_16","alias_value":"NPKJA5VN3H4NDID6","created_at":"2026-08-05T00:46:09.930210+00:00"},{"alias_kind":"pith_short_8","alias_value":"NPKJA5VN","created_at":"2026-08-05T00:46:09.930210+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/NPKJA5VN3H4NDID6D744RVNIP4","json":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4.json","graph_json":"https://pith.science/api/pith-number/NPKJA5VN3H4NDID6D744RVNIP4/graph.json","events_json":"https://pith.science/api/pith-number/NPKJA5VN3H4NDID6D744RVNIP4/events.json","paper":"https://pith.science/paper/NPKJA5VN"},"agent_actions":{"view_html":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4","download_json":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4.json","view_paper":"https://pith.science/paper/NPKJA5VN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.03204&json=true","fetch_graph":"https://pith.science/api/pith-number/NPKJA5VN3H4NDID6D744RVNIP4/graph.json","fetch_events":"https://pith.science/api/pith-number/NPKJA5VN3H4NDID6D744RVNIP4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4/action/storage_attestation","attest_author":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4/action/author_attestation","sign_citation":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4/action/citation_signature","submit_replication":"https://pith.science/pith/NPKJA5VN3H4NDID6D744RVNIP4/action/replication_record"}},"created_at":"2026-08-05T00:46:09.930210+00:00","updated_at":"2026-08-05T00:46:09.930210+00:00"}