LoRA-Based Cascaded Multimodal Fusion for Action Recognition in Healthcare Training Environments

Summary

This paper proposes a cascaded Low-Rank Adaptation (LoRA)-based multimodal fusion framework for action and activity recognition in healthcare-oriented training environments. The architecture combines parameter-efficient modality-specific adaptation with sequential fusion, allowing modalities to be integrated in stages without retraining previously learned components. Instead of assuming a fixed fusion structure, the framework first integrates more closely related modalities and then incorporates additional heterogeneous modalities, enabling scalable adaptation across datasets with different modality sets. The authors evaluate the approach on two healthcare training datasets, NurViD and the Nurse Training dataset. Preliminary results indicate that the cascaded fusion strategy outperforms unimodal models and is competitive against previously reported dataset-specific baselines. The findings suggest that cascaded LoRA fusion is a promising parameter-efficient method for integrating heterogeneous modalities in healthcare training action recognition. Paper: arXiv 2607.11839, by Divya Mereddy and Jeevan Beedareddy.