{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:ZI25MHFILGWBG7BBA77GPZ73KE","short_pith_number":"pith:ZI25MHFI","canonical_record":{"source":{"id":"2605.13161","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T08:24:55Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"beb0232834b0e216947328f2abab2a5263ce1ca2e1638f8761d72953129d16d1","abstract_canon_sha256":"4bc419866106b51d3d0a1e2e64a4efecd75e422a3575cfae6a0cbf0eff7424ad"},"schema_version":"1.0"},"canonical_sha256":"ca35d61ca859ac137c2107fe67e7fb5132ad491dae50852cd8467e86e6cc869f","source":{"kind":"arxiv","id":"2605.13161","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13161","created_at":"2026-05-18T03:08:56Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13161v1","created_at":"2026-05-18T03:08:56Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13161","created_at":"2026-05-18T03:08:56Z"},{"alias_kind":"pith_short_12","alias_value":"ZI25MHFILGWB","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"ZI25MHFILGWBG7BB","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"ZI25MHFI","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:ZI25MHFILGWBG7BBA77GPZ73KE","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13161","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T08:24:55Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"beb0232834b0e216947328f2abab2a5263ce1ca2e1638f8761d72953129d16d1","abstract_canon_sha256":"4bc419866106b51d3d0a1e2e64a4efecd75e422a3575cfae6a0cbf0eff7424ad"},"schema_version":"1.0"},"canonical_sha256":"ca35d61ca859ac137c2107fe67e7fb5132ad491dae50852cd8467e86e6cc869f","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T03:08:56.875034Z","signature_b64":"fbzKl1P33DDwJhCldRxORkVcDXFFnN0VjB4lpf6upnyhXgvUI8VAjNNxFrg8CjJzgm6ZTDQt49qSwix0k6lXCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ca35d61ca859ac137c2107fe67e7fb5132ad491dae50852cd8467e86e6cc869f","last_reissued_at":"2026-05-18T03:08:56.874343Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T03:08:56.874343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13161","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":"2026-05-18T03:08:56Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Pw1iwlNXrhIxBm0V5ZhYbAJBJ6PHEkjidyrQV8kKXp5kRjF9N4G55O+tdtJTt65J9gncI3o6EJdFe/fOYr4aBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T23:28:31.783456Z"},"content_sha256":"a3329babfcabf23fbc1147997beac8037adcf5ff433eb898e74707ae18e506c3","schema_version":"1.0","event_id":"sha256:a3329babfcabf23fbc1147997beac8037adcf5ff433eb898e74707ae18e506c3"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:ZI25MHFILGWBG7BBA77GPZ73KE","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"A$_3$B$_2$: Adaptive Asymmetric Adapter for Alleviating Branch Bias in Vision-Language Image Classification with Few-Shot Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"An adaptive asymmetric adapter suppresses image-branch updates in vision-language models when uncertainty is high, improving few-shot classification.","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Chang Yao, Jingyuan Chen, Kunxi Li, Mingjing Xu, Wenkang Han, Yiyun Zhou, Zhonghua Jiang","submitted_at":"2026-05-13T08:24:55Z","abstract_excerpt":"Efficient transfer learning methods for large-scale vision-language models ($e.g.$, CLIP) enable strong few-shot transfer, yet existing adaptation methods follow a fixed fine-tuning paradigm that implicitly assumes a uniform importance of the image and text branches, which has not been systematically studied in image classification. Through extensive analysis, we reveal a Branch Bias issue in vision-language image classification: adapting the image encoder does not always improve performance under out-of-distribution settings. Motivated by this observation, we propose A$_3$B$_2$, an Adaptive A"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Extensive experiments on three few-shot image classification tasks across 11 datasets demonstrate that A3B2 consistently outperforms 11 competitive prompt- and adapter-based baselines.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that prediction uncertainty reliably indicates when image-branch adaptation should be suppressed, without introducing new failure modes or requiring dataset-specific tuning of the dampening mechanism.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A3B2 adds uncertainty-aware dampening and asymmetric MoE-style adapters to balance image and text branches, outperforming 11 baselines on 11 few-shot datasets.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"An adaptive asymmetric adapter suppresses image-branch updates in vision-language models when uncertainty is high, improving few-shot classification.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"cfe9ec8961d588082e616ddd6eaa142fcd42fbbd166506547510928bd51c731b"},"source":{"id":"2605.13161","kind":"arxiv","version":1},"verdict":{"id":"1f0bdb0c-f613-47b3-93da-fa022575913e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:11:01.039898Z","strongest_claim":"Extensive experiments on three few-shot image classification tasks across 11 datasets demonstrate that A3B2 consistently outperforms 11 competitive prompt- and adapter-based baselines.","one_line_summary":"A3B2 adds uncertainty-aware dampening and asymmetric MoE-style adapters to balance image and text branches, outperforming 11 baselines on 11 few-shot datasets.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that prediction uncertainty reliably indicates when image-branch adaptation should be suppressed, without introducing new failure modes or requiring dataset-specific tuning of the dampening mechanism.","pith_extraction_headline":"An adaptive asymmetric adapter suppresses image-branch updates in vision-language models when uncertainty is high, improving few-shot classification."},"references":{"count":56,"sample":[{"doi":"","year":2022,"title":"Flamingo: a visual language model for few-shot learning.Advances in neural information pro- cessing systems, 35:23716–23736,","work_id":"e8701dea-8ad8-4b3e-9afb-4fb8783ce174","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Food-101–mining discriminative com- ponents with random forests","work_id":"07a32271-7d35-42b1-95cd-178fc8e10e37","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Language models are few-shot learners.Advances in neural information processing sys- tems, 33:1877–1901,","work_id":"cef44151-27de-46af-b76a-7986409ca9eb","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1967,"title":"Markov chains.Springer- Verlag, New York,","work_id":"7bc72276-4640-48b6-925a-398889b12cc8","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2014,"title":"Describing textures in the wild","work_id":"0ab942b6-02b8-49ff-bef1-4bb7a7a5dd8e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":56,"snapshot_sha256":"91d87435f12a519292b4987b4d871709915f3fd8af5502e9f17282fc12890163","internal_anchors":5},"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":"1f0bdb0c-f613-47b3-93da-fa022575913e"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-18T03:08:56Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"1lJI43f+dLHRw28jHdtPpVVT5CgITN6bKpI7gNL2tNL6cyc2Mk6QNkYTczyOzYOojSxy6qliCD5eCYATCTHKAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-28T23:28:31.784534Z"},"content_sha256":"8f878932dded4a0c28bd464e423b7bfb97672050cd8e7affbca8665006ef457c","schema_version":"1.0","event_id":"sha256:8f878932dded4a0c28bd464e423b7bfb97672050cd8e7affbca8665006ef457c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/ZI25MHFILGWBG7BBA77GPZ73KE/bundle.json","state_url":"https://pith.science/pith/ZI25MHFILGWBG7BBA77GPZ73KE/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/ZI25MHFILGWBG7BBA77GPZ73KE/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":"2026-05-28T23:28:31Z","links":{"resolver":"https://pith.science/pith/ZI25MHFILGWBG7BBA77GPZ73KE","bundle":"https://pith.science/pith/ZI25MHFILGWBG7BBA77GPZ73KE/bundle.json","state":"https://pith.science/pith/ZI25MHFILGWBG7BBA77GPZ73KE/state.json","well_known_bundle":"https://pith.science/.well-known/pith/ZI25MHFILGWBG7BBA77GPZ73KE/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:ZI25MHFILGWBG7BBA77GPZ73KE","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":"4bc419866106b51d3d0a1e2e64a4efecd75e422a3575cfae6a0cbf0eff7424ad","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T08:24:55Z","title_canon_sha256":"beb0232834b0e216947328f2abab2a5263ce1ca2e1638f8761d72953129d16d1"},"schema_version":"1.0","source":{"id":"2605.13161","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13161","created_at":"2026-05-18T03:08:56Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13161v1","created_at":"2026-05-18T03:08:56Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13161","created_at":"2026-05-18T03:08:56Z"},{"alias_kind":"pith_short_12","alias_value":"ZI25MHFILGWB","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"ZI25MHFILGWBG7BB","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"ZI25MHFI","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:8f878932dded4a0c28bd464e423b7bfb97672050cd8e7affbca8665006ef457c","target":"graph","created_at":"2026-05-18T03:08:56Z","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 on three few-shot image classification tasks across 11 datasets demonstrate that A3B2 consistently outperforms 11 competitive prompt- and adapter-based baselines."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The assumption that prediction uncertainty reliably indicates when image-branch adaptation should be suppressed, without introducing new failure modes or requiring dataset-specific tuning of the dampening mechanism."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"A3B2 adds uncertainty-aware dampening and asymmetric MoE-style adapters to balance image and text branches, outperforming 11 baselines on 11 few-shot datasets."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"An adaptive asymmetric adapter suppresses image-branch updates in vision-language models when uncertainty is high, improving few-shot classification."}],"snapshot_sha256":"cfe9ec8961d588082e616ddd6eaa142fcd42fbbd166506547510928bd51c731b"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Efficient transfer learning methods for large-scale vision-language models ($e.g.$, CLIP) enable strong few-shot transfer, yet existing adaptation methods follow a fixed fine-tuning paradigm that implicitly assumes a uniform importance of the image and text branches, which has not been systematically studied in image classification. Through extensive analysis, we reveal a Branch Bias issue in vision-language image classification: adapting the image encoder does not always improve performance under out-of-distribution settings. Motivated by this observation, we propose A$_3$B$_2$, an Adaptive A","authors_text":"Chang Yao, Jingyuan Chen, Kunxi Li, Mingjing Xu, Wenkang Han, Yiyun Zhou, Zhonghua Jiang","cross_cats":["cs.LG"],"headline":"An adaptive asymmetric adapter suppresses image-branch updates in vision-language models when uncertainty is high, improving few-shot classification.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T08:24:55Z","title":"A$_3$B$_2$: Adaptive Asymmetric Adapter for Alleviating Branch Bias in Vision-Language Image Classification with Few-Shot Learning"},"references":{"count":56,"internal_anchors":5,"resolved_work":56,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Flamingo: a visual language model for few-shot learning.Advances in neural information pro- cessing systems, 35:23716–23736,","work_id":"e8701dea-8ad8-4b3e-9afb-4fb8783ce174","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Food-101–mining discriminative com- ponents with random forests","work_id":"07a32271-7d35-42b1-95cd-178fc8e10e37","year":2014},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Language models are few-shot learners.Advances in neural information processing sys- tems, 33:1877–1901,","work_id":"cef44151-27de-46af-b76a-7986409ca9eb","year":2020},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Markov chains.Springer- Verlag, New York,","work_id":"7bc72276-4640-48b6-925a-398889b12cc8","year":1967},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Describing textures in the wild","work_id":"0ab942b6-02b8-49ff-bef1-4bb7a7a5dd8e","year":2014}],"snapshot_sha256":"91d87435f12a519292b4987b4d871709915f3fd8af5502e9f17282fc12890163"},"source":{"id":"2605.13161","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-14T19:11:01.039898Z","id":"1f0bdb0c-f613-47b3-93da-fa022575913e","model_set":{"reader":"grok-4.3"},"one_line_summary":"A3B2 adds uncertainty-aware dampening and asymmetric MoE-style adapters to balance image and text branches, outperforming 11 baselines on 11 few-shot datasets.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"An adaptive asymmetric adapter suppresses image-branch updates in vision-language models when uncertainty is high, improving few-shot classification.","strongest_claim":"Extensive experiments on three few-shot image classification tasks across 11 datasets demonstrate that A3B2 consistently outperforms 11 competitive prompt- and adapter-based baselines.","weakest_assumption":"The assumption that prediction uncertainty reliably indicates when image-branch adaptation should be suppressed, without introducing new failure modes or requiring dataset-specific tuning of the dampening mechanism."}},"verdict_id":"1f0bdb0c-f613-47b3-93da-fa022575913e"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:a3329babfcabf23fbc1147997beac8037adcf5ff433eb898e74707ae18e506c3","target":"record","created_at":"2026-05-18T03:08:56Z","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":"4bc419866106b51d3d0a1e2e64a4efecd75e422a3575cfae6a0cbf0eff7424ad","cross_cats_sorted":["cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-13T08:24:55Z","title_canon_sha256":"beb0232834b0e216947328f2abab2a5263ce1ca2e1638f8761d72953129d16d1"},"schema_version":"1.0","source":{"id":"2605.13161","kind":"arxiv","version":1}},"canonical_sha256":"ca35d61ca859ac137c2107fe67e7fb5132ad491dae50852cd8467e86e6cc869f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ca35d61ca859ac137c2107fe67e7fb5132ad491dae50852cd8467e86e6cc869f","first_computed_at":"2026-05-18T03:08:56.874343Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-18T03:08:56.874343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"fbzKl1P33DDwJhCldRxORkVcDXFFnN0VjB4lpf6upnyhXgvUI8VAjNNxFrg8CjJzgm6ZTDQt49qSwix0k6lXCA==","signature_status":"signed_v1","signed_at":"2026-05-18T03:08:56.875034Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13161","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:a3329babfcabf23fbc1147997beac8037adcf5ff433eb898e74707ae18e506c3","sha256:8f878932dded4a0c28bd464e423b7bfb97672050cd8e7affbca8665006ef457c"],"state_sha256":"1685f946fa96f7a42dfcdff2a685c9fb6c75ad10e3bd0ec33ff701f8ef040ae5"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"chEFGmp/0rOh2jVKpA75olIfig58582qeHf/2fJw/FwsCgnI2n8wtEpgVh9EyjmiNBZl6T6rd3e26cAHiBXaAA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-28T23:28:31.789131Z","bundle_sha256":"09458e8a4f5578f365f53b13280ec6bb9ef6b0b00e753beb207fc827b170d9a0"}}