Huyen Le

With Limited Data for Multimodal Alignment, Let the STRUCTURE Guide You

Multimodal models have demonstrated powerful capabilities in complex tasks requiring multimodal alignment, including zero-shot classification and cross-modal retrieval. However, existing models typically rely on millions of paired multimodal samples, which are prohibitively expensive or infeasible t…