Good, Better, Best: Textual Distractors Generation for Multiple-Choice Visual Question Answering via Reinforcement Learning

TL;DR

This paper introduces DG-VQA, a novel unsupervised task for generating challenging distractors in multiple-choice visual question answering, using reinforcement learning with pre-trained VQA models as feedback to improve robustness and data augmentation.

Contribution

The paper proposes Gobbet, a reinforcement learning framework that generates meaningful distractors without ground-truth data, leveraging pre-trained VQA models as an environment for feedback.

Findings

Gobbet effectively generates challenging distractors that fool existing VQA models.

Generated distractors improve model robustness through data augmentation.

Manual analysis reveals factors influencing distractor effectiveness.

Abstract

Multiple-choice VQA has drawn increasing attention from researchers and end-users recently. As the demand for automatically constructing large-scale multiple-choice VQA data grows, we introduce a novel task called textual Distractors Generation for VQA (DG-VQA) focusing on generating challenging yet meaningful distractors given the context image, question, and correct answer. The DG-VQA task aims at generating distractors without ground-truth training samples since such resources are rarely available. To tackle the DG-VQA unsupervisedly, we propose Gobbet, a reinforcement learning(RL) based framework that utilizes pre-trained VQA models as an alternative knowledge base to guide the distractor generation process. In Gobbet, a pre-trained VQA model serves as the environment in RL setting to provide feedback for the input multi-modal query, while a neural distractor generator serves as the agent to take actions accordingly. We propose to use existing VQA models' performance degradation as indicators of the quality of generated distractors. On the other hand, we show the utility of generated distractors through data augmentation experiments, since robustness is more and more important when AI models apply to unpredictable open-domain scenarios or security-sensitive applications. We further conduct a manual case study on the factors why distractors generated by Gobbet can fool existing models.