Visual Question Ansering Using ASP

Abstract

Answer Set Programming (ASP), is a well- known approach to declarative problem solving that has been successfully employed in many application areas. Among them is Visual Question Answering (VQA),which is concerned with answering a question, posed in natural language, about a visual scene shown in an image or possibly also in a video sequence. VQA is a challenging task that requires processing multi-modal input and reasoning capabilities to obtain the correct answer. In this talk, we consider VQA using ASP in a modular neuro- symbolic architecture that comprises both subsymbolic components, based on neural networks, and symbolic reasoning components that use ASP. We discuss explanation finding as a use case, which benefit from the versatility of ASP and the rich landscape of ASP extensions, and briefly touch our ongoing work on exploiting LLMs for VQA.