Alibaba Cloud Developer

Dec 26, 2019 · Artificial Intelligence

How Decomposed Linguistic Representations Overcome Language Priors in VQA

This article reviews a AAAI 2020 paper that introduces a language‑attention based Visual Question Answering model which decomposes questions into type, object, and concept expressions to mitigate language bias, explains its modular architecture, and demonstrates superior performance on VQA‑CP v2 through extensive experiments and ablations.