Counterfactual Visual Explanations

Abstract

In this work, we develop a technique to produce counterfactual visual explanations. Given a 'query' image $I$ for which a vision system predicts class $c$, a counterfactual visual explanation identifies how $I$ could change such that the system would output a different specified class $c^{\prime }$. To do this, we select a 'distractor' image $I^{\prime }$ that the system predicts as class $c^{\prime }$ and identify spatial regions in $I$ and $I^{\prime }$ such that replacing the identified region in $I$ with the identified region in $I^{\prime }$ would push the system towards classifying $I$ as $c^{\prime }$. We apply our approach to multiple image classification datasets generating qualitative results showcasing the interpretability and discriminativeness of our counterfactual explanations. To explore the effectiveness of our explanations in teaching humans, we present machine teaching experiments for the task of fine-grained bird classification. We find that users trained to distinguish bird species fare better when given access to counterfactual explanations in addition to training examples.

References in corpus (5)

• Towards A Rigorous Science of Interpretable Machine Learning
• Sanity Checks for Saliency Maps
• Generating Visual Explanations
• Teaching Categories to Human Learners with Visual Explanations
• Embedding Deep Networks into Visual Explanations

Cited by in corpus (12)

• On Interpretability of Artificial Neural Networks: A Survey
• A survey of algorithmic recourse: definitions, formulations, solutions, and prospects
• Improving Visualization Interpretation Using Counterfactuals
• Fast Real-time Counterfactual Explanations
• Explainable Deep Learning for Uncovering Actionable Scientific Insights for Materials Discovery and Design
• Understanding Learned Reward Functions
• Top-N Recommendation with Counterfactual User Preference Simulation
• Humans learn too: Better Human-AI Interaction using Optimized Human Inputs
• Actionable Attribution Maps for Scientific Machine Learning
• Explaining the Behavior of Black-Box Prediction Algorithms with Causal Learning
• Understanding Misclassifications by Attributes
• Model-Agnostic Explanations using Minimal Forcing Subsets