Despite the advancements in deep learning driven by increased computational power and large datasets, significant challenges remain. These include difficulty in handling novel entities, limited mechanisms for human experts to update knowledge, and lack of interpretability, all of which are crucial for human-centric applications like assistive robotics. To address these issues, we propose leveraging structured information sources, such as knowledge graphs, to enhance the robustness and reliability of deep learning models by utilizing additional domain knowledge. By integrating these knowledge sources through neurosymbolic architectures, which combine neural networks and symbolic reasoning, we can improve model interpretability, generalization, and flexibility. This approach enables AI systems to understand complex scenes and human actions better, ultimately leading to more reliable and transparent performance in real-world scenarios. Our work highlights the potential of augmenting neural networks with additional domain knowledge. Particularly, we demonstrate the benefit of this approach in the task of learning novel objects in a sample-efficient manner and action anticipation from short-video contexts in a human-robot collaborative setting.