Since then, I have been contributing to several research domains including Computer vision, Knowledge Graph Modeling and NLP. In particular, in a recent paper ("Classification by Attention"), we showed that the injection of commonsense to scene representations leads to a significantly large improvement in label efficiency. In general, I'm interested in understanding the fascinating underlying concepts of human intelligence and employing those concepts to improve the current AI models. My current area of research lies between representation learning, graphs, cognitive models, and NLP. My interest in interdisciplinary research sometimes leads to collaborations with biologists, physicists, robotic engineers, etc.