Improving Visual Question Answering Models through Robustness Analysis and In-Context Learning with a Chain of Basic Questions

Jia-Hong Huang, Modar Alfadly, Bernard Ghanem +1

Deep neural networks have been critical in the task of Visual Question Answering (VQA), with research traditionally focused on improving model accuracy. Recently, however, there ha…