Learning Abstract Visual Reasoning via Task Decomposition: A Case Study in Raven Progressive Matrices

TL;DR

This paper introduces a transformer-based deep learning model that decomposes abstract visual reasoning tasks in Raven Progressive Matrices into subgoals, predicting object properties to improve reasoning and reduce bias.

Contribution

The novel approach predicts visual object properties and arrangements to solve RPMs, outperforming existing methods and offering interpretability and bias mitigation.

Findings

Model outperforms state-of-the-art methods.

Provides insights into inference process.

Reduces known benchmark biases.

Abstract

Learning to perform abstract reasoning often requires decomposing the task in question into intermediate subgoals that are not specified upfront, but need to be autonomously devised by the learner. In Raven Progressive Matrices (RPM), the task is to choose one of the available answers given a context, where both the context and answers are composite images featuring multiple objects in various spatial arrangements. As this high-level goal is the only guidance available, learning to solve RPMs is challenging. In this study, we propose a deep learning architecture based on the transformer blueprint which, rather than directly making the above choice, addresses the subgoal of predicting the visual properties of individual objects and their arrangements. The multidimensional predictions obtained in this way are then directly juxtaposed to choose the answer. We consider a few ways in which the model parses the visual input into tokens and several regimes of masking parts of the input in self-supervised training. In experimental assessment, the models not only outperform state-of-the-art methods but also provide interesting insights and partial explanations about the inference. The design of the method also makes it immune to biases that are known to be present in some RPM benchmarks.