# Noisy Data Training

ALIGN is a vision-language dual-encoder model that aligns image and text representations through contrastive learning, achieving state-of-the-art cross-modal representation with large-scale noisy data.