Aligning by Misaligning: Boundary-aware Curriculum Learning for Multimodal Alignment
This work improves multimodal alignment for tasks like image-text retrieval by handling borderline cases, offering a domain-specific advancement.
The paper tackles the problem of multimodal alignment by addressing ambiguous negative pairs that differ only slightly from positives, proposing a lightweight add-on method that achieves up to +32% R@1 over CLIP and sets new SOTA on four large-scale benchmarks without extra labels.
Most multimodal models treat every negative pair alike, ignoring the ambiguous negatives that differ from the positive by only a small detail. We propose Boundary-Aware Curriculum with Local Attention (BACL), a lightweight add-on that turns these borderline cases into a curriculum signal. A Boundary-aware Negative Sampler gradually raises difficulty, while a Contrastive Local Attention loss highlights where the mismatch occurs. The two modules are fully differentiable and work with any off-the-shelf dual encoder. Theory predicts a fast O(1/n) error rate; practice shows up to +32% R@1 over CLIP and new SOTA on four large-scale benchmarks, all without extra labels.