{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3HC5PCGOQ225K22U6HMDKULVX3","short_pith_number":"pith:3HC5PCGO","schema_version":"1.0","canonical_sha256":"d9c5d788ce86b5d56b54f1d8355175becc5d3c4e31c80eeeb5652901539c16f2","source":{"kind":"arxiv","id":"2605.31457","version":1},"attestation_state":"computed","paper":{"title":"VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hengbo Xu, Shengjie Jin, Yanbiao Ma, Zhiwu Lu","submitted_at":"2026-05-29T15:51:12Z","abstract_excerpt":"With the rapid advancement of large multimodal models (LMMs), inference-time overhead has become a key bottleneck for real-world deployment. Existing methods typically prune visual tokens at prefill, assuming the required visual evidence remains static during reasoning. However, we empirically show that visual evidence is strongly step-dependent: only a sparse subset of visual tokens is critical at each decoding step, and the critical set evolves across reasoning. Furthermore, we identify a coupled bottleneck where redundant visual context can steer the model toward query-irrelevant regions, l"},"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":"2605.31457","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-29T15:51:12Z","cross_cats_sorted":[],"title_canon_sha256":"6ffd3aea9878b02e2e6d7119ccfb24dcd0fbfcfb6964ec2e7f2eb02aacea77a5","abstract_canon_sha256":"53ab3841554be5a4f70f2f56a48248306297c9c5e3f97250999746f68de566e3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-01T02:04:06.868053Z","signature_b64":"5tqG7HKeokpkBSC75ugT/W9bu+TrvwqBlBcTBaoOvdiF3m0n9UrG+uxpbF9TXqwMUjFXLCVDPN7hWIMKkm5ZDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d9c5d788ce86b5d56b54f1d8355175becc5d3c4e31c80eeeb5652901539c16f2","last_reissued_at":"2026-06-01T02:04:06.867200Z","signature_status":"signed_v1","first_computed_at":"2026-06-01T02:04:06.867200Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hengbo Xu, Shengjie Jin, Yanbiao Ma, Zhiwu Lu","submitted_at":"2026-05-29T15:51:12Z","abstract_excerpt":"With the rapid advancement of large multimodal models (LMMs), inference-time overhead has become a key bottleneck for real-world deployment. Existing methods typically prune visual tokens at prefill, assuming the required visual evidence remains static during reasoning. However, we empirically show that visual evidence is strongly step-dependent: only a sparse subset of visual tokens is critical at each decoding step, and the critical set evolves across reasoning. Furthermore, we identify a coupled bottleneck where redundant visual context can steer the model toward query-irrelevant regions, l"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.31457","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/2605.31457/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":"2605.31457","created_at":"2026-06-01T02:04:06.867350+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.31457v1","created_at":"2026-06-01T02:04:06.867350+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.31457","created_at":"2026-06-01T02:04:06.867350+00:00"},{"alias_kind":"pith_short_12","alias_value":"3HC5PCGOQ225","created_at":"2026-06-01T02:04:06.867350+00:00"},{"alias_kind":"pith_short_16","alias_value":"3HC5PCGOQ225K22U","created_at":"2026-06-01T02:04:06.867350+00:00"},{"alias_kind":"pith_short_8","alias_value":"3HC5PCGO","created_at":"2026-06-01T02:04:06.867350+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/3HC5PCGOQ225K22U6HMDKULVX3","json":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3.json","graph_json":"https://pith.science/api/pith-number/3HC5PCGOQ225K22U6HMDKULVX3/graph.json","events_json":"https://pith.science/api/pith-number/3HC5PCGOQ225K22U6HMDKULVX3/events.json","paper":"https://pith.science/paper/3HC5PCGO"},"agent_actions":{"view_html":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3","download_json":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3.json","view_paper":"https://pith.science/paper/3HC5PCGO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.31457&json=true","fetch_graph":"https://pith.science/api/pith-number/3HC5PCGOQ225K22U6HMDKULVX3/graph.json","fetch_events":"https://pith.science/api/pith-number/3HC5PCGOQ225K22U6HMDKULVX3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3/action/storage_attestation","attest_author":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3/action/author_attestation","sign_citation":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3/action/citation_signature","submit_replication":"https://pith.science/pith/3HC5PCGOQ225K22U6HMDKULVX3/action/replication_record"}},"created_at":"2026-06-01T02:04:06.867350+00:00","updated_at":"2026-06-01T02:04:06.867350+00:00"}