{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HKIK7EMP3XQ2PUC25AE3H26PUC","short_pith_number":"pith:HKIK7EMP","schema_version":"1.0","canonical_sha256":"3a90af918fdde1a7d05ae809b3ebcfa0a06f5faf9e806000468895117dcea8d2","source":{"kind":"arxiv","id":"2605.15735","version":2},"attestation_state":"computed","paper":{"title":"UAM: A Dual-Stream Perspective on Forgetting in VLA Training","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hongbin Xu, Jianke Zhang, Jianyu Chen, Tian Lan, Xiaoyu Chen, Yanjiang Guo, Yuanfei Luo, Yucheng Hu, Ziyang Liu","submitted_at":"2026-05-15T08:45:37Z","abstract_excerpt":"Vision--language--action (VLA) models are typically built by fine-tuning a pretrained vision--language model (VLM) on action data. However, we show that this standard recipe systematically erodes the VLM's multimodal competence, a side effect we call the embodiment tax. But do VLAs have to forget? Inspired by the two-stream organization of biological vision, we trace this degradation to a structural bottleneck: current VLAs ask a single encoder to support both language-grounded semantics and control-relevant visual features, whereas biological vision separates recognition and visuomotor contro"},"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":"2605.15735","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-15T08:45:37Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c17cb96fe3caa252142af322a234f45bf3b0931584957a0d0863dddbeb1f706c","abstract_canon_sha256":"c79e9ce24803fa462447e2b9cf7b700d424d3029e7d7a537e88028ad8f9640f7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:38.428720Z","signature_b64":"be3YGjRYdo1kKO4Hs/bGmGjGl2kEoEt8u+aNhJ5mW2veJyCKS07TWGdheCxh+riB7AXdQTufbxsiMdslZ8xlBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3a90af918fdde1a7d05ae809b3ebcfa0a06f5faf9e806000468895117dcea8d2","last_reissued_at":"2026-05-20T00:04:38.427703Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:38.427703Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"UAM: A Dual-Stream Perspective on Forgetting in VLA Training","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Hongbin Xu, Jianke Zhang, Jianyu Chen, Tian Lan, Xiaoyu Chen, Yanjiang Guo, Yuanfei Luo, Yucheng Hu, Ziyang Liu","submitted_at":"2026-05-15T08:45:37Z","abstract_excerpt":"Vision--language--action (VLA) models are typically built by fine-tuning a pretrained vision--language model (VLM) on action data. However, we show that this standard recipe systematically erodes the VLM's multimodal competence, a side effect we call the embodiment tax. But do VLAs have to forget? Inspired by the two-stream organization of biological vision, we trace this degradation to a structural bottleneck: current VLAs ask a single encoder to support both language-grounded semantics and control-relevant visual features, whereas biological vision separates recognition and visuomotor contro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.15735","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/2605.15735/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":"2605.15735","created_at":"2026-05-20T00:04:38.427859+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15735v2","created_at":"2026-05-20T00:04:38.427859+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15735","created_at":"2026-05-20T00:04:38.427859+00:00"},{"alias_kind":"pith_short_12","alias_value":"HKIK7EMP3XQ2","created_at":"2026-05-20T00:04:38.427859+00:00"},{"alias_kind":"pith_short_16","alias_value":"HKIK7EMP3XQ2PUC2","created_at":"2026-05-20T00:04:38.427859+00:00"},{"alias_kind":"pith_short_8","alias_value":"HKIK7EMP","created_at":"2026-05-20T00:04:38.427859+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/HKIK7EMP3XQ2PUC25AE3H26PUC","json":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC.json","graph_json":"https://pith.science/api/pith-number/HKIK7EMP3XQ2PUC25AE3H26PUC/graph.json","events_json":"https://pith.science/api/pith-number/HKIK7EMP3XQ2PUC25AE3H26PUC/events.json","paper":"https://pith.science/paper/HKIK7EMP"},"agent_actions":{"view_html":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC","download_json":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC.json","view_paper":"https://pith.science/paper/HKIK7EMP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15735&json=true","fetch_graph":"https://pith.science/api/pith-number/HKIK7EMP3XQ2PUC25AE3H26PUC/graph.json","fetch_events":"https://pith.science/api/pith-number/HKIK7EMP3XQ2PUC25AE3H26PUC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC/action/storage_attestation","attest_author":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC/action/author_attestation","sign_citation":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC/action/citation_signature","submit_replication":"https://pith.science/pith/HKIK7EMP3XQ2PUC25AE3H26PUC/action/replication_record"}},"created_at":"2026-05-20T00:04:38.427859+00:00","updated_at":"2026-05-20T00:04:38.427859+00:00"}