{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZMQPNYINLA4KW3XS6KP63F3BWA","short_pith_number":"pith:ZMQPNYIN","schema_version":"1.0","canonical_sha256":"cb20f6e10d5838ab6ef2f29fed9761b037fa85e20b9072ceb5ae721304035009","source":{"kind":"arxiv","id":"2606.24156","version":1},"attestation_state":"computed","paper":{"title":"Accelerating Multimodal Large Language Models with Prior-Corrected Token Reduction","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jianwei Yin, Jingcai Guo, Taotao Cai, Yuxiang Cai, Zengjie Chen, Zhi Chen","submitted_at":"2026-06-23T05:24:50Z","abstract_excerpt":"Visual token reduction has emerged as an effective strategy for accelerating Multimodal Large Language Models (MLLMs). Many existing methods prune tokens by ranking text-visual attention scores. However, we show that attention is often dominated by a model-induced prior: even without textual instruction, MLLMs tend to focus on certain task-agnostic regions. Consequently, the attention scores of instruction-conditioned tokens are suppressed, increasing the risk that these tokens are discarded during pruning. To address this issue, we propose Prior-Corrected Token Reduction (PriorTR), a training"},"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":"2606.24156","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-06-23T05:24:50Z","cross_cats_sorted":[],"title_canon_sha256":"c04d89b784190e77e09aad7a123779612be3a90fcabafdd4800d6e9d3b190a7e","abstract_canon_sha256":"63ecbbad1105abd54c4fa40877bb076841ec621098eb46a75d2ded5aba3242f9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-24T01:14:43.137185Z","signature_b64":"EVHV3QaLBCdpqke9VvFjWvLubUIIq7+Tlbe4RRtC/J30DVXIK0pRKmi7mjQfCg1kl765Bt+OtfAZGpNrazT0CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cb20f6e10d5838ab6ef2f29fed9761b037fa85e20b9072ceb5ae721304035009","last_reissued_at":"2026-06-24T01:14:43.136271Z","signature_status":"signed_v1","first_computed_at":"2026-06-24T01:14:43.136271Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Accelerating Multimodal Large Language Models with Prior-Corrected Token Reduction","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jianwei Yin, Jingcai Guo, Taotao Cai, Yuxiang Cai, Zengjie Chen, Zhi Chen","submitted_at":"2026-06-23T05:24:50Z","abstract_excerpt":"Visual token reduction has emerged as an effective strategy for accelerating Multimodal Large Language Models (MLLMs). Many existing methods prune tokens by ranking text-visual attention scores. However, we show that attention is often dominated by a model-induced prior: even without textual instruction, MLLMs tend to focus on certain task-agnostic regions. Consequently, the attention scores of instruction-conditioned tokens are suppressed, increasing the risk that these tokens are discarded during pruning. To address this issue, we propose Prior-Corrected Token Reduction (PriorTR), a training"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.24156","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/2606.24156/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":"2606.24156","created_at":"2026-06-24T01:14:43.136451+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.24156v1","created_at":"2026-06-24T01:14:43.136451+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.24156","created_at":"2026-06-24T01:14:43.136451+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZMQPNYINLA4K","created_at":"2026-06-24T01:14:43.136451+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZMQPNYINLA4KW3XS","created_at":"2026-06-24T01:14:43.136451+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZMQPNYIN","created_at":"2026-06-24T01:14:43.136451+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/ZMQPNYINLA4KW3XS6KP63F3BWA","json":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA.json","graph_json":"https://pith.science/api/pith-number/ZMQPNYINLA4KW3XS6KP63F3BWA/graph.json","events_json":"https://pith.science/api/pith-number/ZMQPNYINLA4KW3XS6KP63F3BWA/events.json","paper":"https://pith.science/paper/ZMQPNYIN"},"agent_actions":{"view_html":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA","download_json":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA.json","view_paper":"https://pith.science/paper/ZMQPNYIN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.24156&json=true","fetch_graph":"https://pith.science/api/pith-number/ZMQPNYINLA4KW3XS6KP63F3BWA/graph.json","fetch_events":"https://pith.science/api/pith-number/ZMQPNYINLA4KW3XS6KP63F3BWA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA/action/storage_attestation","attest_author":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA/action/author_attestation","sign_citation":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA/action/citation_signature","submit_replication":"https://pith.science/pith/ZMQPNYINLA4KW3XS6KP63F3BWA/action/replication_record"}},"created_at":"2026-06-24T01:14:43.136451+00:00","updated_at":"2026-06-24T01:14:43.136451+00:00"}