{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TJSS3BR26LREWICEGWZZPLPJSB","short_pith_number":"pith:TJSS3BR2","schema_version":"1.0","canonical_sha256":"9a652d863af2e24b204435b397ade99054e6923a4cd4fd94e7920479bf8e380a","source":{"kind":"arxiv","id":"2502.02562","version":1},"attestation_state":"computed","paper":{"title":"Learning the RoPEs: Better 2D and 3D Position Encodings with STRING","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.RO","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alex Bewley, Alexey Gritsenko, Arunkumar Byravan, Avinava Dubey, Ayzaan Wahid, Chuyuan Fu, Connor Schenck, David Rendleman, Deepali Jain, Dmitry Kalashnikov, Isaac Reid, Jake Varley, Jonathan Tompson, Joshua Ainslie, Krzysztof Choromanski, Matthias Minderer, Mithun George Jacob, Mohit Sharma, Ren\\'e Wagner, Sumeet Singh, Tianli Ding, Vikas Sindhwani","submitted_at":"2025-02-04T18:37:17Z","abstract_excerpt":"We introduce STRING: Separable Translationally Invariant Position Encodings. STRING extends Rotary Position Encodings, a recently proposed and widely used algorithm in large language models, via a unifying theoretical framework. Importantly, STRING still provides exact translation invariance, including token coordinates of arbitrary dimensionality, whilst maintaining a low computational footprint. These properties are especially important in robotics, where efficient 3D token representation is key. We integrate STRING into Vision Transformers with RGB(-D) inputs (color plus optional depth), sh"},"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":"2502.02562","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-04T18:37:17Z","cross_cats_sorted":["cs.AI","cs.CV","cs.RO","stat.ML"],"title_canon_sha256":"f2e57c8b723e13a7721ded47f1133c6388d0f664f9f074a0b222671064e62ff9","abstract_canon_sha256":"a43de47e899da1808630ee4f1c7869351b0e2c55021de79a029bf07c6af2423d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:09:53.101470Z","signature_b64":"P+edvs9g3iGEpXgyQmuLSwKWPUkQWLLmSDC6e2dhA7rYMFwzu+6jeanBDUt9c51ULnNTBWd/lkY8YbkJjRyBCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a652d863af2e24b204435b397ade99054e6923a4cd4fd94e7920479bf8e380a","last_reissued_at":"2026-07-05T10:09:53.101051Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:09:53.101051Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning the RoPEs: Better 2D and 3D Position Encodings with STRING","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.RO","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alex Bewley, Alexey Gritsenko, Arunkumar Byravan, Avinava Dubey, Ayzaan Wahid, Chuyuan Fu, Connor Schenck, David Rendleman, Deepali Jain, Dmitry Kalashnikov, Isaac Reid, Jake Varley, Jonathan Tompson, Joshua Ainslie, Krzysztof Choromanski, Matthias Minderer, Mithun George Jacob, Mohit Sharma, Ren\\'e Wagner, Sumeet Singh, Tianli Ding, Vikas Sindhwani","submitted_at":"2025-02-04T18:37:17Z","abstract_excerpt":"We introduce STRING: Separable Translationally Invariant Position Encodings. STRING extends Rotary Position Encodings, a recently proposed and widely used algorithm in large language models, via a unifying theoretical framework. Importantly, STRING still provides exact translation invariance, including token coordinates of arbitrary dimensionality, whilst maintaining a low computational footprint. These properties are especially important in robotics, where efficient 3D token representation is key. We integrate STRING into Vision Transformers with RGB(-D) inputs (color plus optional depth), sh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.02562","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/2502.02562/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":"2502.02562","created_at":"2026-07-05T10:09:53.101103+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.02562v1","created_at":"2026-07-05T10:09:53.101103+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.02562","created_at":"2026-07-05T10:09:53.101103+00:00"},{"alias_kind":"pith_short_12","alias_value":"TJSS3BR26LRE","created_at":"2026-07-05T10:09:53.101103+00:00"},{"alias_kind":"pith_short_16","alias_value":"TJSS3BR26LREWICE","created_at":"2026-07-05T10:09:53.101103+00:00"},{"alias_kind":"pith_short_8","alias_value":"TJSS3BR2","created_at":"2026-07-05T10:09:53.101103+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2512.07805","citing_title":"Group Representational Position Encoding","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2602.00937","citing_title":"CLAMP: Contrastive Learning for 3D Multi-View Action-Conditioned Robotic Manipulation Pretraining","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12491","citing_title":"Elastic Attention Cores for Scalable Vision Transformers","ref_index":157,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB","json":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB.json","graph_json":"https://pith.science/api/pith-number/TJSS3BR26LREWICEGWZZPLPJSB/graph.json","events_json":"https://pith.science/api/pith-number/TJSS3BR26LREWICEGWZZPLPJSB/events.json","paper":"https://pith.science/paper/TJSS3BR2"},"agent_actions":{"view_html":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB","download_json":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB.json","view_paper":"https://pith.science/paper/TJSS3BR2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.02562&json=true","fetch_graph":"https://pith.science/api/pith-number/TJSS3BR26LREWICEGWZZPLPJSB/graph.json","fetch_events":"https://pith.science/api/pith-number/TJSS3BR26LREWICEGWZZPLPJSB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB/action/storage_attestation","attest_author":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB/action/author_attestation","sign_citation":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB/action/citation_signature","submit_replication":"https://pith.science/pith/TJSS3BR26LREWICEGWZZPLPJSB/action/replication_record"}},"created_at":"2026-07-05T10:09:53.101103+00:00","updated_at":"2026-07-05T10:09:53.101103+00:00"}