Measuring CLEVRness: Blackbox testing of Visual Reasoning Models

TL;DR

This paper introduces a behavioral testing framework for visual reasoning models using a game where an adversary reconfigures scenes to test if models truly reason or exploit dataset biases, revealing limitations in current models.

Contribution

It proposes a novel black-box testing method involving an adversarial scene reconfiguration to evaluate reasoning abilities of visual QA models.

Findings

CLEVR models can be easily fooled by adversarial reconfigurations

Current models may rely on dataset biases rather than true reasoning

The method provides a controlled way to measure reasoning efficiency

Abstract

How can we measure the reasoning capabilities of intelligence systems? Visual question answering provides a convenient framework for testing the model's abilities by interrogating the model through questions about the scene. However, despite scores of various visual QA datasets and architectures, which sometimes yield even a super-human performance, the question of whether those architectures can actually reason remains open to debate. To answer this, we extend the visual question answering framework and propose the following behavioral test in the form of a two-player game. We consider black-box neural models of CLEVR. These models are trained on a diagnostic dataset benchmarking reasoning. Next, we train an adversarial player that re-configures the scene to fool the CLEVR model. We show that CLEVR models, which otherwise could perform at a human level, can easily be fooled by our agent. Our results put in doubt whether data-driven approaches can do reasoning without exploiting the numerous biases that are often present in those datasets. Finally, we also propose a controlled experiment measuring the efficiency of such models to learn and perform reasoning.