Interact before Align: Leveraging Cross-Modal Knowledge for Domain Adaptive Action Recognition | IEEE Conference Publication | IEEE Xplore