{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:U6J2OD54ACYHPFBJWAECFPES4P","short_pith_number":"pith:U6J2OD54","canonical_record":{"source":{"id":"2605.13835","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T17:56:23Z","cross_cats_sorted":[],"title_canon_sha256":"bbe520d2b68ca1ed89b41ab3c8ced9b1702690ee969a576c1c0529a747da49d6","abstract_canon_sha256":"c26a1c0cad34eea2175be1335d6ebfaed5362e077320d5b34f78c4ef05e8ab4f"},"schema_version":"1.0"},"canonical_sha256":"a793a70fbc00b0779429b00822bc92e3e181b1c1a3ef2a907fb50e73c0b47cc2","source":{"kind":"arxiv","id":"2605.13835","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13835","created_at":"2026-05-18T02:44:14Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13835v1","created_at":"2026-05-18T02:44:14Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13835","created_at":"2026-05-18T02:44:14Z"},{"alias_kind":"pith_short_12","alias_value":"U6J2OD54ACYH","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"U6J2OD54ACYHPFBJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"U6J2OD54","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:U6J2OD54ACYHPFBJWAECFPES4P","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13835","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T17:56:23Z","cross_cats_sorted":[],"title_canon_sha256":"bbe520d2b68ca1ed89b41ab3c8ced9b1702690ee969a576c1c0529a747da49d6","abstract_canon_sha256":"c26a1c0cad34eea2175be1335d6ebfaed5362e077320d5b34f78c4ef05e8ab4f"},"schema_version":"1.0"},"canonical_sha256":"a793a70fbc00b0779429b00822bc92e3e181b1c1a3ef2a907fb50e73c0b47cc2","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:14.945301Z","signature_b64":"F4mfkc4BYiSOF0bhBBaad9doArHcHCyS7dSIBX3fAylDZrKE84WTgdYnFmIpq606Miv9Q3KkE4GN5JtnFRgfCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a793a70fbc00b0779429b00822bc92e3e181b1c1a3ef2a907fb50e73c0b47cc2","last_reissued_at":"2026-05-18T02:44:14.944708Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:14.944708Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13835","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:14Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"K4cyx4PB32LoxmciLlHS2vnUsVyDlKM1NJRmX0Ps4PQsNK6bV89rFco9Nc3ZF9IFUF9f//I3CsroA+ujN22fAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T16:11:53.048720Z"},"content_sha256":"f026c510c5bb3faf40d5170d6874f1f6cece6d7b407e2f8f3fef8f770ceb5e50","schema_version":"1.0","event_id":"sha256:f026c510c5bb3faf40d5170d6874f1f6cece6d7b407e2f8f3fef8f770ceb5e50"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:U6J2OD54ACYHPFBJWAECFPES4P","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Unlocking Patch-Level Features for CLIP-Based Class-Incremental Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Aligning CLIP patch features to semantic descriptions improves class-incremental learning.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Da-Wei Zhou, Hao Sun, Zi-Jun Ding","submitted_at":"2026-05-13T17:56:23Z","abstract_excerpt":"Class-Incremental Learning (CIL) enables models to continuously integrate new knowledge while mitigating catastrophic forgetting. Driven by the remarkable generalization of CLIP, leveraging pre-trained vision-language models has become a dominant paradigm in CIL. However, current work primarily focuses on aligning global image embeddings (i.e., [CLS] token) with their corresponding text prompts (i.e., [EOS] token). Despite their good performance, we find that they discard the rich patch-level semantic information inherent in CLIP's encoders. For instance, when recognizing a rabbit, local patch"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Extensive experiments demonstrate that SPA achieves state-of-the-art performance.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That GPT-5-generated class-wise semantic descriptions reliably identify discriminative patches and that optimal transport alignment between selected patches and semantic tokens yields a meaningful recognition improvement beyond global embeddings.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SPA unlocks patch-level features in CLIP for class-incremental learning via semantic-guided selection and optimal transport alignment with class descriptions, plus projectors and pseudo-feature replay to reduce forgetting.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Aligning CLIP patch features to semantic descriptions improves class-incremental learning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"27bdb55f10c22e42961089c59148ccdacf9ac3bec2f98e727ce5917e37b6accd"},"source":{"id":"2605.13835","kind":"arxiv","version":1},"verdict":{"id":"248d67c7-6a84-4771-8a88-be4cb1b92b19","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:08:58.877996Z","strongest_claim":"Extensive experiments demonstrate that SPA achieves state-of-the-art performance.","one_line_summary":"SPA unlocks patch-level features in CLIP for class-incremental learning via semantic-guided selection and optimal transport alignment with class descriptions, plus projectors and pseudo-feature replay to reduce forgetting.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That GPT-5-generated class-wise semantic descriptions reliably identify discriminative patches and that optimal transport alignment between selected patches and semantic tokens yields a meaningful recognition improvement beyond global embeddings.","pith_extraction_headline":"Aligning CLIP patch features to semantic descriptions improves class-incremental learning."},"references":{"count":72,"sample":[{"doi":"","year":2018,"title":"Memory aware synapses: Learning what (not) to forget","work_id":"0925aaf0-e501-443e-a639-34de9b7a3620","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Qwen-vl: A versatile vision-language model for understanding, localization.Text Reading, and Beyond, 2(1):1, 2023","work_id":"3e704b76-920c-4ff8-85b2-7603f3c5ddd3","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Objectnet: A large-scale bias-controlled dataset for pushing the limits of object recognition models.Advances in neural information processing systems, 32, 2019","work_id":"1e1c2889-d7d1-4115-a84d-eb33cf223329","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Food-101–mining discriminative components with random forests","work_id":"5fc3ab66-8c71-4e08-9abb-b01e46165194","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Efficient lifelong learning with A-GEM.CoRR, abs/1812.00420","work_id":"c1e3aa81-39b5-44a3-a1b5-6fa343e0b956","ref_index":5,"cited_arxiv_id":"1812.00420","is_internal_anchor":true}],"resolved_work":72,"snapshot_sha256":"b6bfcde75964b003cec5e88700e1db4610c5422d61988370c0d03a15bc560de8","internal_anchors":7},"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"},"verdict_id":"248d67c7-6a84-4771-8a88-be4cb1b92b19"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T02:44:14Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"fiymF4Sl58Lpp4Q3jsSBoF009xWhxwsq+5BbD4Of0LBQYv9is/DSVgyKjJChy3Mf++9+wFVM1TDEhnLZXYe+Dw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-26T16:11:53.049817Z"},"content_sha256":"b582351cbd053c58d58d4dbc50918349085db1772b2ae1f15e80e42ea78736bb","schema_version":"1.0","event_id":"sha256:b582351cbd053c58d58d4dbc50918349085db1772b2ae1f15e80e42ea78736bb"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/U6J2OD54ACYHPFBJWAECFPES4P/bundle.json","state_url":"https://pith.science/pith/U6J2OD54ACYHPFBJWAECFPES4P/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/U6J2OD54ACYHPFBJWAECFPES4P/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-26T16:11:53Z","links":{"resolver":"https://pith.science/pith/U6J2OD54ACYHPFBJWAECFPES4P","bundle":"https://pith.science/pith/U6J2OD54ACYHPFBJWAECFPES4P/bundle.json","state":"https://pith.science/pith/U6J2OD54ACYHPFBJWAECFPES4P/state.json","well_known_bundle":"https://pith.science/.well-known/pith/U6J2OD54ACYHPFBJWAECFPES4P/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:U6J2OD54ACYHPFBJWAECFPES4P","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"c26a1c0cad34eea2175be1335d6ebfaed5362e077320d5b34f78c4ef05e8ab4f","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T17:56:23Z","title_canon_sha256":"bbe520d2b68ca1ed89b41ab3c8ced9b1702690ee969a576c1c0529a747da49d6"},"schema_version":"1.0","source":{"id":"2605.13835","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13835","created_at":"2026-05-18T02:44:14Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13835v1","created_at":"2026-05-18T02:44:14Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13835","created_at":"2026-05-18T02:44:14Z"},{"alias_kind":"pith_short_12","alias_value":"U6J2OD54ACYH","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"U6J2OD54ACYHPFBJ","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"U6J2OD54","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:b582351cbd053c58d58d4dbc50918349085db1772b2ae1f15e80e42ea78736bb","target":"graph","created_at":"2026-05-18T02:44:14Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Extensive experiments demonstrate that SPA achieves state-of-the-art performance."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That GPT-5-generated class-wise semantic descriptions reliably identify discriminative patches and that optimal transport alignment between selected patches and semantic tokens yields a meaningful recognition improvement beyond global embeddings."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SPA unlocks patch-level features in CLIP for class-incremental learning via semantic-guided selection and optimal transport alignment with class descriptions, plus projectors and pseudo-feature replay to reduce forgetting."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Aligning CLIP patch features to semantic descriptions improves class-incremental learning."}],"snapshot_sha256":"27bdb55f10c22e42961089c59148ccdacf9ac3bec2f98e727ce5917e37b6accd"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Class-Incremental Learning (CIL) enables models to continuously integrate new knowledge while mitigating catastrophic forgetting. Driven by the remarkable generalization of CLIP, leveraging pre-trained vision-language models has become a dominant paradigm in CIL. However, current work primarily focuses on aligning global image embeddings (i.e., [CLS] token) with their corresponding text prompts (i.e., [EOS] token). Despite their good performance, we find that they discard the rich patch-level semantic information inherent in CLIP's encoders. For instance, when recognizing a rabbit, local patch","authors_text":"Da-Wei Zhou, Hao Sun, Zi-Jun Ding","cross_cats":[],"headline":"Aligning CLIP patch features to semantic descriptions improves class-incremental learning.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T17:56:23Z","title":"Unlocking Patch-Level Features for CLIP-Based Class-Incremental Learning"},"references":{"count":72,"internal_anchors":7,"resolved_work":72,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Memory aware synapses: Learning what (not) to forget","work_id":"0925aaf0-e501-443e-a639-34de9b7a3620","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Qwen-vl: A versatile vision-language model for understanding, localization.Text Reading, and Beyond, 2(1):1, 2023","work_id":"3e704b76-920c-4ff8-85b2-7603f3c5ddd3","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Objectnet: A large-scale bias-controlled dataset for pushing the limits of object recognition models.Advances in neural information processing systems, 32, 2019","work_id":"1e1c2889-d7d1-4115-a84d-eb33cf223329","year":2019},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Food-101–mining discriminative components with random forests","work_id":"5fc3ab66-8c71-4e08-9abb-b01e46165194","year":2014},{"cited_arxiv_id":"1812.00420","doi":"","is_internal_anchor":true,"ref_index":5,"title":"Efficient lifelong learning with A-GEM.CoRR, abs/1812.00420","work_id":"c1e3aa81-39b5-44a3-a1b5-6fa343e0b956","year":2018}],"snapshot_sha256":"b6bfcde75964b003cec5e88700e1db4610c5422d61988370c0d03a15bc560de8"},"source":{"id":"2605.13835","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T19:08:58.877996Z","id":"248d67c7-6a84-4771-8a88-be4cb1b92b19","model_set":{"reader":"grok-4.3"},"one_line_summary":"SPA unlocks patch-level features in CLIP for class-incremental learning via semantic-guided selection and optimal transport alignment with class descriptions, plus projectors and pseudo-feature replay to reduce forgetting.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Aligning CLIP patch features to semantic descriptions improves class-incremental learning.","strongest_claim":"Extensive experiments demonstrate that SPA achieves state-of-the-art performance.","weakest_assumption":"That GPT-5-generated class-wise semantic descriptions reliably identify discriminative patches and that optimal transport alignment between selected patches and semantic tokens yields a meaningful recognition improvement beyond global embeddings."}},"verdict_id":"248d67c7-6a84-4771-8a88-be4cb1b92b19"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f026c510c5bb3faf40d5170d6874f1f6cece6d7b407e2f8f3fef8f770ceb5e50","target":"record","created_at":"2026-05-18T02:44:14Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"c26a1c0cad34eea2175be1335d6ebfaed5362e077320d5b34f78c4ef05e8ab4f","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T17:56:23Z","title_canon_sha256":"bbe520d2b68ca1ed89b41ab3c8ced9b1702690ee969a576c1c0529a747da49d6"},"schema_version":"1.0","source":{"id":"2605.13835","kind":"arxiv","version":1}},"canonical_sha256":"a793a70fbc00b0779429b00822bc92e3e181b1c1a3ef2a907fb50e73c0b47cc2","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a793a70fbc00b0779429b00822bc92e3e181b1c1a3ef2a907fb50e73c0b47cc2","first_computed_at":"2026-05-18T02:44:14.944708Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T02:44:14.944708Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"F4mfkc4BYiSOF0bhBBaad9doArHcHCyS7dSIBX3fAylDZrKE84WTgdYnFmIpq606Miv9Q3KkE4GN5JtnFRgfCQ==","signature_status":"signed_v1","signed_at":"2026-05-18T02:44:14.945301Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13835","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f026c510c5bb3faf40d5170d6874f1f6cece6d7b407e2f8f3fef8f770ceb5e50","sha256:b582351cbd053c58d58d4dbc50918349085db1772b2ae1f15e80e42ea78736bb"],"state_sha256":"7c4e21a7350d4b28040801928e87b7287dd6ace196f8b988776fd062e4ebb15a"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"tkK557ouUOy2lIxLk793ywgZ7bCPrURZvhnlwCPrD6kS7snaBees+3WK0uQx0VfWKqWC6VLOwEIZMjLUQvBFCA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-26T16:11:53.054339Z","bundle_sha256":"4ab3362d69e2e245f9344f1371505964aee26ac5bfd3f49b6e55bf2cbf5c5aca"}}