Generative Counterfactual Introspection for Explainable Deep Learning

By Shusen Liu, Bhavya Kailkhura, Donald Loveland, Yong Han. 2019

97 citations recorded by the source corpus.

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.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Francisco Herrera
2. Salvador Garcia
3. Daniel Molina
4. Christian Walter Peter Omlin
5. Waddah Saeed
6. Javier Del Ser
7. Divyat Mahajan
8. Ruocheng Guo

In Computer Vision and Pattern Recognition

1. Salvador Garcia
2. Francisco Herrera
3. Raha Moraffah
4. Divyat Mahajan
5. Adrien Bennetot
6. Alejandro Barredo Arrieta
7. Sergio Gil-Lopez
8. Martin Pereira-Fariña

In Safety, Robustness & Interpretability

1. Divyat Mahajan
2. Chenhao Tan
3. Amit Sharma
4. Raha Moraffah
5. Adrien Bennetot
6. Martin Pereira-Fariña
7. Mansooreh Karami
8. Ilia Stepin

Citations in the corpus

• Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI (2019, positive): “Another interesting research direction is the use of generative models for the creation of counterfactuals, i.e., modifications to the input data that could eventually alter the original prediction of the model [368] .”
• Causal Interpretability for Machine Learning - Problems, Methods and Evaluation (2020, methodology): “the counterfactual input xcf and the desired counterfactual output, while the second term indicates the distance between the actual instance features x and the counterfactual features xcf. Liu et al. [60] propose a generative model to generate counterfactual explanations for explaining a model’s dec”
• A Survey of Contrastive and Counterfactual Explanation Generation Methods for Explainable Artificial Intelligence (2021, background): “…[159], gradient boosted decision trees [147], support vector machines [131], [138], [146], random forests [81], [86], [142]–[144], neural networks [6], [48], [49], [91], [129], [130], [133], [135], [139], [141], [145], [148], [151], or combinations of these [100], [105], [134], [152], [154], [160].”
• Preserving Causal Constraints in Counterfactual Explanations for Machine Learning Classifiers (2019, methodology): “Liu et al. [11] proposes the use of Generative Adversarial Networks [6] as a generative framework for counterfactuals.”
• Explainable AI (XAI): A Systematic Meta-Survey of Current Challenges and Future Opportunities (2021, background): “The second is creating counterfactuals through generative models [115].”