CMAP: Cross-Modal Adaptive Prompting for Multi-Domain Task-Incremental Learning
Researchers have developed CMAP, a novel method for multi-domain task-incremental learning that leverages cross-modal text embeddings from CLIP. Unlike previous approaches that solely relied on visual features for task routing and adaptation, CMAP utilizes text-space task routing and symmetric cross-modal gating. This new technique achieves state-of-the-art performance on the MTIL benchmark, outperforming existing methods by significant margins with a minimal number of trainable parameters. AI
IMPACT This research advances task-incremental learning by effectively utilizing cross-modal information, potentially leading to more robust and efficient AI models in diverse, sequential learning scenarios.