Generative Counterfactual Introspection for Explainable Deep Learning
arXiv:1907.03077
Abstract
In this work, we propose an introspection technique for deep neural networks that relies on a generative model to instigate salient editing of the input image for model interpretation. Such modification provides the fundamental interventional operation that allows us to obtain answers to counterfactual inquiries, i.e., what meaningful change can be made to the input image in order to alter the prediction. We demonstrate how to reveal interesting properties of the given classifiers by utilizing the proposed introspection approach on both the MNIST and the CelebA dataset.
References in corpus (7)
- Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
- Understanding Neural Networks Through Deep Visualization
- Searching for Exotic Particles in High-Energy Physics with Deep Learning
- Counterfactual Fairness
- Multifaceted Feature Visualization: Uncovering the Different Types of Features Learned By Each Neuron in Deep Neural Networks
- TreeView: Peeking into Deep Neural Networks Via Feature-Space Partitioning
- Explaining Deep Learning Models using Causal Inference
Cited by in corpus (11)
- Socially Responsible AI Algorithms: Issues, Purposes, and Challenges
- Explanation by Progressive Exaggeration
- Causal Interpretability for Machine Learning -- Problems, Methods and Evaluation
- Explainable Deep Learning for Uncovering Actionable Scientific Insights for Materials Discovery and Design
- Generative Counterfactuals for Neural Networks via Attribute-Informed Perturbation
- SCOUT: Self-aware Discriminant Counterfactual Explanations
- Actionable Attribution Maps for Scientific Machine Learning
- Reliable Graph Neural Network Explanations Through Adversarial Training
- Causal Learning for Socially Responsible AI
- Discriminative Attribution from Counterfactuals
- DA-DGCEx: Ensuring Validity of Deep Guided Counterfactual Explanations With Distribution-Aware Autoencoder Loss