Understanding the computational demands underlying visual reasoning

TL;DR

This paper investigates the computational complexity of visual reasoning by testing CNNs on a set of visual reasoning tasks, extending models with attention mechanisms, and proposing a taxonomy based on relation types and complexity.

Contribution

It introduces a taxonomy of visual reasoning tasks based on relation types and complexity, and demonstrates that attention-augmented CNNs improve performance on these tasks.

Findings

Attention mechanisms significantly improve CNN performance on complex visual reasoning tasks.

The taxonomy links task difficulty to relation type and number of relations.

Attention networks partially explain the proposed taxonomy.

Abstract

Visual understanding requires comprehending complex visual relations between objects within a scene. Here, we seek to characterize the computational demands for abstract visual reasoning. We do this by systematically assessing the ability of modern deep convolutional neural networks (CNNs) to learn to solve the "Synthetic Visual Reasoning Test" (SVRT) challenge, a collection of twenty-three visual reasoning problems. Our analysis reveals a novel taxonomy of visual reasoning tasks, which can be primarily explained by both the type of relations (same-different vs. spatial-relation judgments) and the number of relations used to compose the underlying rules. Prior cognitive neuroscience work suggests that attention plays a key role in humans' visual reasoning ability. To test this hypothesis, we extended the CNNs with spatial and feature-based attention mechanisms. In a second series of experiments, we evaluated the ability of these attention networks to learn to solve the SVRT challenge and found the resulting architectures to be much more efficient at solving the hardest of these visual reasoning tasks. Most importantly, the corresponding improvements on individual tasks partially explained our novel taxonomy. Overall, this work provides an granular computational account of visual reasoning and yields testable neuroscience predictions regarding the differential need for feature-based vs. spatial attention depending on the type of visual reasoning problem.