{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:FN6NJHTG3WSCCCDY7MKG72DC4Z","short_pith_number":"pith:FN6NJHTG","schema_version":"1.0","canonical_sha256":"2b7cd49e66dda4210878fb146fe862e67b00c30a68a1dd93c57a71d950eb6814","source":{"kind":"arxiv","id":"2506.10967","version":2},"attestation_state":"computed","paper":{"title":"Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Junwen Pan, Lichen Li, Mengzhen Liu, Ming Lu, Qi She, Qizhe Zhang, Shanghang Zhang, Yuan Zhang","submitted_at":"2025-06-12T17:59:09Z","abstract_excerpt":"In multimodal large language models (MLLMs), the length of input visual tokens is often significantly greater than that of their textual counterparts, leading to a high inference cost. Many works aim to address this issue by removing redundant visual tokens. However, current approaches either rely on attention-based pruning, which retains numerous duplicate tokens, or use similarity-based pruning, overlooking the instruction relevance, consequently causing suboptimal performance. In this paper, we go beyond attention or similarity by proposing a novel visual token pruning method named CDPruner"},"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":"2506.10967","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-06-12T17:59:09Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c580bb0d424802dc98af6f5d1818477e5f946f420ffd03f09fc43edc4e61e0ed","abstract_canon_sha256":"4df179919323a0be0ca501d984e87bbbb4b7ff6e3923a8e8ab4321b0fd250db1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:30:01.727948Z","signature_b64":"7KD0YTYlgolKYXDiV9dXfnh39VVm4ZbO+hTonweMKqrUBIvU1NLfPEadqrxdaKlzelitHNqxzOuq6zKn1tZMDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2b7cd49e66dda4210878fb146fe862e67b00c30a68a1dd93c57a71d950eb6814","last_reissued_at":"2026-07-05T11:30:01.727535Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:30:01.727535Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond Attention or Similarity: Maximizing Conditional Diversity for Token Pruning in MLLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Junwen Pan, Lichen Li, Mengzhen Liu, Ming Lu, Qi She, Qizhe Zhang, Shanghang Zhang, Yuan Zhang","submitted_at":"2025-06-12T17:59:09Z","abstract_excerpt":"In multimodal large language models (MLLMs), the length of input visual tokens is often significantly greater than that of their textual counterparts, leading to a high inference cost. Many works aim to address this issue by removing redundant visual tokens. However, current approaches either rely on attention-based pruning, which retains numerous duplicate tokens, or use similarity-based pruning, overlooking the instruction relevance, consequently causing suboptimal performance. In this paper, we go beyond attention or similarity by proposing a novel visual token pruning method named CDPruner"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.10967","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/2506.10967/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":"2506.10967","created_at":"2026-07-05T11:30:01.727590+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.10967v2","created_at":"2026-07-05T11:30:01.727590+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.10967","created_at":"2026-07-05T11:30:01.727590+00:00"},{"alias_kind":"pith_short_12","alias_value":"FN6NJHTG3WSC","created_at":"2026-07-05T11:30:01.727590+00:00"},{"alias_kind":"pith_short_16","alias_value":"FN6NJHTG3WSCCCDY","created_at":"2026-07-05T11:30:01.727590+00:00"},{"alias_kind":"pith_short_8","alias_value":"FN6NJHTG","created_at":"2026-07-05T11:30:01.727590+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":13,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24165","citing_title":"Spectral Evolution-Guided Token Pruning in Multimodal Large Language Models","ref_index":86,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27161","citing_title":"TOPS: First-Principles Visual Token Pruning via Constructing Token Optimal Preservation Sets for Efficient MLLM Inference","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2607.02484","citing_title":"Combating Textual Noise and Redundancy: Entropy-Aware Dense Visual Token Pruning","ref_index":72,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03075","citing_title":"TGV-KV: Text-Grounded KV Eviction for Vision-Language Models","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02482","citing_title":"X-Stream: Exploring MLLMs as Multiplexers for Multi-Stream Understanding","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02482","citing_title":"X-Stream: Exploring MLLMs as Multiplexers for Multi-Stream Understanding","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29662","citing_title":"SAFE-Pruner: Semantic Attention-Guided Future-Aware Token Pruning for Efficient Vision-Language-Action Manipulation","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00392","citing_title":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20950","citing_title":"Focus-then-Context: Subject-Centric Progressive Visual Token Reduction for Vision-Language Models","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00392","citing_title":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00392","citing_title":"RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07812","citing_title":"HAWK: Head Importance-Aware Visual Token Pruning in Multimodal Models","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20937","citing_title":"Sink-Token-Aware Pruning for Fine-Grained Video Understanding in Efficient Video LLMs","ref_index":52,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z","json":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z.json","graph_json":"https://pith.science/api/pith-number/FN6NJHTG3WSCCCDY7MKG72DC4Z/graph.json","events_json":"https://pith.science/api/pith-number/FN6NJHTG3WSCCCDY7MKG72DC4Z/events.json","paper":"https://pith.science/paper/FN6NJHTG"},"agent_actions":{"view_html":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z","download_json":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z.json","view_paper":"https://pith.science/paper/FN6NJHTG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.10967&json=true","fetch_graph":"https://pith.science/api/pith-number/FN6NJHTG3WSCCCDY7MKG72DC4Z/graph.json","fetch_events":"https://pith.science/api/pith-number/FN6NJHTG3WSCCCDY7MKG72DC4Z/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z/action/storage_attestation","attest_author":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z/action/author_attestation","sign_citation":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z/action/citation_signature","submit_replication":"https://pith.science/pith/FN6NJHTG3WSCCCDY7MKG72DC4Z/action/replication_record"}},"created_at":"2026-07-05T11:30:01.727590+00:00","updated_at":"2026-07-05T11:30:01.727590+00:00"}