Variance-Aware Loss Scheduling for Multimodal Alignment in Low-Data Settings

Sneh Pillai

Training vision-language models for image-text alignment typically requires large datasets to achieve robust performance. In low-data scenarios, standard contrastive learning can s…