AI RESEARCH
Label What Matters: Modality-Balanced and Difficulty-Aware Multimodal Active Learning
arXiv CS.CV
•
ArXi:2603.25107v1 Announce Type: new Multimodal learning integrates complementary information from different modalities such as image, text, and audio to improve model performance, but its success relies on large-scale labeled data, which is costly to obtain. Active learning (AL) mitigates this challenge by selectively annotating informative samples.