EmergentBridge: Improving Zero-Shot Cross-Modal Transfer in Unified Multimodal Embedding Models

In three linesEmergentBridge improves unified multimodal embedding models for unpaired modality pairs (audio↔depth, infrared↔audio). The method learns a mapping producing a 'noisy bridge anchor' and enforces alignment in the orthogonal subspace, preserving existing anchor-alignment structure. Results across 9 datasets: outperforms baselines on zero-shot classification and retrieval.

Summary generated by Claude — human-verified