Ctrl+K
- claim-1-rovtl-robust-vision-tabular-learning-is-proposed-in-section-3-as-a-contrastive-vision-tabular-pretraining-method-designed-to-remain-robust-across-the-full-range-of-tabular-data-availability-from-0-to-100-of-tabular-attributes-present-section-3
- claim-2-rovtl-introduces-random-masking-of-tabular-attributes-simulating-attribute-level-missingness-as-a-data-augmentation-during-contrastive-pretraining-to-make-the-learned-joint-representation-robust-to-missing-tabular-values-at-test-time-section-3
- claim-3-section-5-evaluates-rovtl-against-baseline-vision-tabular-methods-on-uk-biobank-cardiac-mri-data-reporting-comparisons-in-table-3-across-varying-rates-of-tabular-missingness-section-5-table-3
- claim-4-an-ablation-isolating-the-contribution-of-the-missingness-augmentation-strategy-versus-standard-contrastive-pretraining-is-reported-evaluating-its-effect-on-downstream-robustness-table-6
- claim-5-additional-experiments-on-the-german-and-time-based-tabular-datasets-cited-as-german-et-al-2014-and-dugdale-et-al-1999-are-used-alongside-uk-biobank-to-assess-generalization-of-rovtl-s-robustness-to-missing-tabular-data-section-4-section-5
- conclusion
- executive-summary
- repro-no-data-no-problem-robust-vision-tabular-learning-with-missing-values
- 3.3 kB