{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EJGYZS4YZHSNQLCFQUKIPCW5UO","short_pith_number":"pith:EJGYZS4Y","schema_version":"1.0","canonical_sha256":"224d8ccb98c9e4d82c458514878adda3a85b149af462fe7d3a51c78f78ce1091","source":{"kind":"arxiv","id":"2507.19839","version":1},"attestation_state":"computed","paper":{"title":"GNSP: Gradient Null Space Projection for Preserving Cross-Modal Alignment in VLMs Continual Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Qiuhe Hong, Shuo Yang, Tiantian Peng, Yonghong Tian, Yuyang Liu","submitted_at":"2025-07-26T07:22:12Z","abstract_excerpt":"Contrastive Language-Image Pretraining has demonstrated remarkable zero-shot generalization by aligning visual and textual modalities in a shared embedding space. However, when continuously fine-tuned on diverse tasks, CLIP suffers from catastrophic forgetting and degradation of its embedding alignment, undermining its zero-shot capabilities. In this work, we propose Gradient Null Space Projection (GNSP), an efficient continual learning method that projects task-specific gradients onto the null space of previously learned knowledge. This orthogonal projection mathematically prevents interferen"},"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":"2507.19839","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-07-26T07:22:12Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"4a667dcca55d2402dd18d3ec72380fa32fe6e68e6f38c3be7a7606d17586d687","abstract_canon_sha256":"b5251b675aa10f75c12d499215d1f5793edbb4668e7c4f3a9587f6290306ddec"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:43:43.005587Z","signature_b64":"a/I9C1PUBGekVhGOgKJ4YMez5RXcz+uHF7TH5kyJP2XLdQwlZ9FQGQy6XKkwIMd/QLWQbmTQMMgLWl0IeRuoBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"224d8ccb98c9e4d82c458514878adda3a85b149af462fe7d3a51c78f78ce1091","last_reissued_at":"2026-07-05T11:43:43.005159Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:43:43.005159Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GNSP: Gradient Null Space Projection for Preserving Cross-Modal Alignment in VLMs Continual Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Qiuhe Hong, Shuo Yang, Tiantian Peng, Yonghong Tian, Yuyang Liu","submitted_at":"2025-07-26T07:22:12Z","abstract_excerpt":"Contrastive Language-Image Pretraining has demonstrated remarkable zero-shot generalization by aligning visual and textual modalities in a shared embedding space. However, when continuously fine-tuned on diverse tasks, CLIP suffers from catastrophic forgetting and degradation of its embedding alignment, undermining its zero-shot capabilities. In this work, we propose Gradient Null Space Projection (GNSP), an efficient continual learning method that projects task-specific gradients onto the null space of previously learned knowledge. This orthogonal projection mathematically prevents interferen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.19839","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/2507.19839/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":"2507.19839","created_at":"2026-07-05T11:43:43.005220+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.19839v1","created_at":"2026-07-05T11:43:43.005220+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.19839","created_at":"2026-07-05T11:43:43.005220+00:00"},{"alias_kind":"pith_short_12","alias_value":"EJGYZS4YZHSN","created_at":"2026-07-05T11:43:43.005220+00:00"},{"alias_kind":"pith_short_16","alias_value":"EJGYZS4YZHSNQLCF","created_at":"2026-07-05T11:43:43.005220+00:00"},{"alias_kind":"pith_short_8","alias_value":"EJGYZS4Y","created_at":"2026-07-05T11:43:43.005220+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2508.04227","citing_title":"Continual Learning for VLMs: A Survey and Taxonomy Beyond Forgetting","ref_index":97,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19301","citing_title":"iGSP:Implicit Gradient Subspace Projection for Efficient Continual Learning of Vision-Language Models","ref_index":45,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO","json":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO.json","graph_json":"https://pith.science/api/pith-number/EJGYZS4YZHSNQLCFQUKIPCW5UO/graph.json","events_json":"https://pith.science/api/pith-number/EJGYZS4YZHSNQLCFQUKIPCW5UO/events.json","paper":"https://pith.science/paper/EJGYZS4Y"},"agent_actions":{"view_html":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO","download_json":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO.json","view_paper":"https://pith.science/paper/EJGYZS4Y","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.19839&json=true","fetch_graph":"https://pith.science/api/pith-number/EJGYZS4YZHSNQLCFQUKIPCW5UO/graph.json","fetch_events":"https://pith.science/api/pith-number/EJGYZS4YZHSNQLCFQUKIPCW5UO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO/action/storage_attestation","attest_author":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO/action/author_attestation","sign_citation":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO/action/citation_signature","submit_replication":"https://pith.science/pith/EJGYZS4YZHSNQLCFQUKIPCW5UO/action/replication_record"}},"created_at":"2026-07-05T11:43:43.005220+00:00","updated_at":"2026-07-05T11:43:43.005220+00:00"}