{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:KUTWRTSSJJGPSRLAEW2X4VUN3L","short_pith_number":"pith:KUTWRTSS","schema_version":"1.0","canonical_sha256":"552768ce524a4cf9456025b57e568ddaf1662cb51ef547cece2c86dc9d61fe7d","source":{"kind":"arxiv","id":"2202.12015","version":1},"attestation_state":"computed","paper":{"title":"Learning to Merge Tokens in Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Andr\\'e Susano Pinto, Basil Mustafa, Carlos Riquelme, Cedric Renggli, Joan Puigcerver, Neil Houlsby","submitted_at":"2022-02-24T10:56:17Z","abstract_excerpt":"Transformers are widely applied to solve natural language understanding and computer vision tasks. While scaling up these architectures leads to improved performance, it often comes at the expense of much higher computational costs. In order for large-scale models to remain practical in real-world systems, there is a need for reducing their computational overhead. In this work, we present the PatchMerger, a simple module that reduces the number of patches or tokens the network has to process by merging them between two consecutive intermediate layers. We show that the PatchMerger achieves a si"},"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":"2202.12015","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2022-02-24T10:56:17Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"2f1c4960dfbd5860ff3bd6db8a61d6fab46e55bd443e93d9a86814ff6279b248","abstract_canon_sha256":"2c34745d781541c008ca70f342fbce14935a1e96b8fc17d653a84624c2054cbc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:59:48.951723Z","signature_b64":"k81CKic5zAheajvpIOR/2lTyksJDIWfs2Bw6iktvuge2nowR1lRBoU6Ta6k+OS9F0qE7Si0vgMoaZzaMPBXaBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"552768ce524a4cf9456025b57e568ddaf1662cb51ef547cece2c86dc9d61fe7d","last_reissued_at":"2026-07-05T03:59:48.951176Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:59:48.951176Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning to Merge Tokens in Vision Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Andr\\'e Susano Pinto, Basil Mustafa, Carlos Riquelme, Cedric Renggli, Joan Puigcerver, Neil Houlsby","submitted_at":"2022-02-24T10:56:17Z","abstract_excerpt":"Transformers are widely applied to solve natural language understanding and computer vision tasks. While scaling up these architectures leads to improved performance, it often comes at the expense of much higher computational costs. In order for large-scale models to remain practical in real-world systems, there is a need for reducing their computational overhead. In this work, we present the PatchMerger, a simple module that reduces the number of patches or tokens the network has to process by merging them between two consecutive intermediate layers. We show that the PatchMerger achieves a si"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2202.12015","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/2202.12015/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":"2202.12015","created_at":"2026-07-05T03:59:48.951237+00:00"},{"alias_kind":"arxiv_version","alias_value":"2202.12015v1","created_at":"2026-07-05T03:59:48.951237+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2202.12015","created_at":"2026-07-05T03:59:48.951237+00:00"},{"alias_kind":"pith_short_12","alias_value":"KUTWRTSSJJGP","created_at":"2026-07-05T03:59:48.951237+00:00"},{"alias_kind":"pith_short_16","alias_value":"KUTWRTSSJJGPSRLA","created_at":"2026-07-05T03:59:48.951237+00:00"},{"alias_kind":"pith_short_8","alias_value":"KUTWRTSS","created_at":"2026-07-05T03:59:48.951237+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.19953","citing_title":"ConsisFormer: Compute-Efficient Transformer for Wireless Foundation Models Based on Channel Consistency","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19849","citing_title":"ViCoStream: Streaming VideoLLMs Can Run Beyond 100 FPS with Stage-Wise Coordinated Inference","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31676","citing_title":"REDI: Corpus Aware Patch Ranking for DINOv3 Token Reduction","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29350","citing_title":"Fast Enough to Act: Spatio-Temporal Visual Token Merging for Low-Latency Robotic VLMs and VLAs","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2509.02560","citing_title":"FastVGGT: Training-Free Acceleration of Visual Geometry Transformer","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14563","citing_title":"Revisiting Token Compression for Accelerating ViT-based Sparse Multi-View 3D Object Detectors","ref_index":37,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L","json":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L.json","graph_json":"https://pith.science/api/pith-number/KUTWRTSSJJGPSRLAEW2X4VUN3L/graph.json","events_json":"https://pith.science/api/pith-number/KUTWRTSSJJGPSRLAEW2X4VUN3L/events.json","paper":"https://pith.science/paper/KUTWRTSS"},"agent_actions":{"view_html":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L","download_json":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L.json","view_paper":"https://pith.science/paper/KUTWRTSS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2202.12015&json=true","fetch_graph":"https://pith.science/api/pith-number/KUTWRTSSJJGPSRLAEW2X4VUN3L/graph.json","fetch_events":"https://pith.science/api/pith-number/KUTWRTSSJJGPSRLAEW2X4VUN3L/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L/action/storage_attestation","attest_author":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L/action/author_attestation","sign_citation":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L/action/citation_signature","submit_replication":"https://pith.science/pith/KUTWRTSSJJGPSRLAEW2X4VUN3L/action/replication_record"}},"created_at":"2026-07-05T03:59:48.951237+00:00","updated_at":"2026-07-05T03:59:48.951237+00:00"}