SARN: Relational Reasoning through Sequential Attention

TL;DR

This paper introduces SARN, an attention-augmented relational network that enhances relational reasoning efficiency and accuracy by selectively pairing objects, significantly reducing computational costs and improving performance on relational tasks.

Contribution

SARN is a novel attention-based relational network that improves efficiency and accuracy in relational reasoning by focusing on relevant object pairs, reducing computational complexity.

Findings

Achieves high accuracy on Sort-of-CLEVR dataset

Reduces computational and memory requirements

Excels particularly on relational questions

Abstract

This paper proposes an attention module augmented relational network called SARN(Sequential Attention Relational Network) that can carry out relational reasoning by extracting reference objects and making efficient pairing between objects. SARN greatly reduces the computational and memory requirements of the relational network, which computes all object pairs. It also shows high accuracy on the Sort-of-CLEVR dataset compared to other models, especially on relational questions.