{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:MO6DIRVTZGQATPDKGX3PI73CQN","short_pith_number":"pith:MO6DIRVT","schema_version":"1.0","canonical_sha256":"63bc3446b3c9a009bc6a35f6f47f628347985ff07e293027f5912c03f3b0e9bc","source":{"kind":"arxiv","id":"2603.20020","version":2},"attestation_state":"computed","paper":{"title":"Detached Skip-Links and $R$-Probe: Decoupling Feature Aggregation from Gradient Propagation for MLLM OCR","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Chengxin Zheng, Daxiang Dong, Ming Zhang, Ruchang Yao, Yusheng Zhao, Ziye Yuan","submitted_at":"2026-03-20T15:07:14Z","abstract_excerpt":"Multimodal large language models (MLLMs) excel at high-level reasoning yet fail on OCR tasks where fine-grained visual details are compromised or misaligned. We identify an overlooked optimization issue in multi-layer feature fusion. Skip pathways introduce direct back-propagation paths from high-level semantic objectives to early visual layers. This mechanism overwrites low-level signals and destabilizes training. To mitigate this gradient interference, we propose Detached Skip-Links, a minimal modification that reuses shallow features in the forward pass while stopping gradients through the "},"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":"2603.20020","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-03-20T15:07:14Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a376c6a1f5ecb3e1454569f850ac6cc56cbfab4d63bdfd8aaade90a9fd843a5d","abstract_canon_sha256":"ac2ada5139b2aa04ec114e9aa609fbb8811fae0cbf8651898f08b09416a08e21"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:05:47.243097Z","signature_b64":"JCbREDdd20CoIvfFiWgp1s4OLnd8xk6xX4iGM+SeEIt4sKgLI5Xl6c62wRd1dwPgjAx9ZeNjl/VjdZnoSiRwCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"63bc3446b3c9a009bc6a35f6f47f628347985ff07e293027f5912c03f3b0e9bc","last_reissued_at":"2026-05-27T01:05:47.242330Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:05:47.242330Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Detached Skip-Links and $R$-Probe: Decoupling Feature Aggregation from Gradient Propagation for MLLM OCR","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Chengxin Zheng, Daxiang Dong, Ming Zhang, Ruchang Yao, Yusheng Zhao, Ziye Yuan","submitted_at":"2026-03-20T15:07:14Z","abstract_excerpt":"Multimodal large language models (MLLMs) excel at high-level reasoning yet fail on OCR tasks where fine-grained visual details are compromised or misaligned. We identify an overlooked optimization issue in multi-layer feature fusion. Skip pathways introduce direct back-propagation paths from high-level semantic objectives to early visual layers. This mechanism overwrites low-level signals and destabilizes training. To mitigate this gradient interference, we propose Detached Skip-Links, a minimal modification that reuses shallow features in the forward pass while stopping gradients through the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.20020","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/2603.20020/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":"2603.20020","created_at":"2026-05-27T01:05:47.242466+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.20020v2","created_at":"2026-05-27T01:05:47.242466+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.20020","created_at":"2026-05-27T01:05:47.242466+00:00"},{"alias_kind":"pith_short_12","alias_value":"MO6DIRVTZGQA","created_at":"2026-05-27T01:05:47.242466+00:00"},{"alias_kind":"pith_short_16","alias_value":"MO6DIRVTZGQATPDK","created_at":"2026-05-27T01:05:47.242466+00:00"},{"alias_kind":"pith_short_8","alias_value":"MO6DIRVT","created_at":"2026-05-27T01:05:47.242466+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/MO6DIRVTZGQATPDKGX3PI73CQN","json":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN.json","graph_json":"https://pith.science/api/pith-number/MO6DIRVTZGQATPDKGX3PI73CQN/graph.json","events_json":"https://pith.science/api/pith-number/MO6DIRVTZGQATPDKGX3PI73CQN/events.json","paper":"https://pith.science/paper/MO6DIRVT"},"agent_actions":{"view_html":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN","download_json":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN.json","view_paper":"https://pith.science/paper/MO6DIRVT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.20020&json=true","fetch_graph":"https://pith.science/api/pith-number/MO6DIRVTZGQATPDKGX3PI73CQN/graph.json","fetch_events":"https://pith.science/api/pith-number/MO6DIRVTZGQATPDKGX3PI73CQN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN/action/storage_attestation","attest_author":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN/action/author_attestation","sign_citation":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN/action/citation_signature","submit_replication":"https://pith.science/pith/MO6DIRVTZGQATPDKGX3PI73CQN/action/replication_record"}},"created_at":"2026-05-27T01:05:47.242466+00:00","updated_at":"2026-05-27T01:05:47.242466+00:00"}