{"paper":{"title":"SMA: Submodular Modality Aligner For Data Efficient Multimodal Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SMA aligns images and text by optimizing submodular mutual information over sets of descriptions rather than individual pairs, enabling strong zero-shot performance with only tens of thousands of samples.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Anay Majee, Rishabh Iyer, Truong Pham","submitted_at":"2026-05-13T01:36:43Z","abstract_excerpt":"Despite the recent success of Multimodal Foundation Models (FMs), their reliance on massive paired datasets limits their applicability in low-data and rare-scenario settings where aligned data is scarce and expensive. A key bottleneck is the adoption of an instance-level formulation, which learns alignment by maximizing correlation between individual image-text pairs while neglecting the underlying geometric structure across modalities resulting in a modality gap across input modalities. In this paper, we propose a combinatorial paradigm for multimodal alignment that moves beyond pairwise lear"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"SMA achieves strong multimodal generalization using only tens of thousands of samples. This is orders of magnitude fewer than standard approaches.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the set-based formulation with submodular mutual information captures richer cross-modal geometric structure and effectively utilizes multiple positive associations without introducing biases or requiring extensive post-hoc tuning that affects the reported gains.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SMA uses a submodular mutual information objective on data sets to deliver competitive zero-shot classification and retrieval performance on CLIP benchmarks with only tens of thousands of samples, orders of magnitude fewer than standard approaches.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SMA aligns images and text by optimizing submodular mutual information over sets of descriptions rather than individual pairs, enabling strong zero-shot performance with only tens of thousands of samples.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4e1704971da14415c7088e0a800e567c6c9ecf21f815d7685ecb04c04fdccea9"},"source":{"id":"2605.12872","kind":"arxiv","version":1},"verdict":{"id":"cca2a51a-d698-480c-944c-647333612a5b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:28:21.138946Z","strongest_claim":"SMA achieves strong multimodal generalization using only tens of thousands of samples. This is orders of magnitude fewer than standard approaches.","one_line_summary":"SMA uses a submodular mutual information objective on data sets to deliver competitive zero-shot classification and retrieval performance on CLIP benchmarks with only tens of thousands of samples, orders of magnitude fewer than standard approaches.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the set-based formulation with submodular mutual information captures richer cross-modal geometric structure and effectively utilizes multiple positive associations without introducing biases or requiring extensive post-hoc tuning that affects the reported gains.","pith_extraction_headline":"SMA aligns images and text by optimizing submodular mutual information over sets of descriptions rather than individual pairs, enabling strong zero-shot performance with only tens of thousands of samples."},"references":{"count":54,"sample":[{"doi":"","year":2026,"title":"Liteembed: Adapting clip to rare classes, 2026","work_id":"797ef698-5fd1-44b3-a38c-c6748e0bc479","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Theoretical analysis of submodular information measures for targeted data subset selection.ArXiv, abs/2402.13454, 2024","work_id":"8bdfc0a7-ae56-4245-bdf7-5394f7e823f8","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"arXiv preprint arXiv:2106.15324 , year=","work_id":"9e312208-7e3a-4663-ab37-1d9cc653c348","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Submodularity in machine learning and artificial intelligence","work_id":"83a2d723-7834-4f2a-a012-e49eae74b5c5","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2002,"title":"A Simple Framework for Contrastive Learning of Visual Representations","work_id":"77d995ce-c44e-4692-9c54-cf8ce771464a","ref_index":5,"cited_arxiv_id":"2002.05709","is_internal_anchor":true}],"resolved_work":54,"snapshot_sha256":"e71b506350c44a0670cbbd0109066f8555609e414a443a92e123ae756e78777e","internal_anchors":5},"formal_canon":{"evidence_count":2,"snapshot_sha256":"4d4512cc3b4b824f742f870a1403d8342de3aae4e79768480a484bc57141d2d4"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}