Explanations based on the Missing: Towards Contrastive Explanations with Pertinent Negatives

By Amit Dhurandhar, Pin-Yu Chen, Ronny Luss, Chun-Chen Tu, Pai-Shun Ting, Karthikeyan Shanmugam, Payel Das. 2018

687 citations recorded by the source corpus.

In this paper we propose a novel method that provides contrastive explanations justifying the classification of an input by a black box classifier such as a deep neural network. Given an input we find what should be %necessarily and minimally and sufficiently present (viz. important object pixels in an image) to justify its classification and analogously what should be minimally and necessarily \emph{absent} (viz. certain background pixels). We argue that such explanations are natural for humans and are used commonly in domains such as health care and criminology. What is minimally but critically \emph{absent} is an important part of an explanation, which to the best of our knowledge, has not been explicitly identified by current explanation methods that explain predictions of neural networks. We validate our approach on three real datasets obtained from diverse domains; namely, a handwritten digits dataset MNIST, a large procurement fraud dataset and a brain activity strength dataset. In all three cases, we witness the power of our approach in generating precise explanations that are also easy for human experts to understand and evaluate.

Subfields: Artificial Intelligence

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

Citations in the corpus

Getting a CLUE: A Method for Explaining Uncertainty Estimates (2020, background): “[21] Amit Dhurandhar, Pin-Yu Chen, Ronny Luss, Chun-Chen Tu, Paishun Ting, Karthikeyan Shanmugam, and Payel Das.”
Counterfactual Visual Explanations (2019, background): “More relevant to our approach, Dhurandhar et al. (2018) ﬁnd minimal regions in the input image which should be necessarily present/absent for a particular classiﬁcation.”
A survey of algorithmic recourse: definitions, formulations, solutions, and prospects (2020, methodology): “focuses on likely states in the (empirical) distribution of features [52, 53, 87, 89, 104, 143] identifying instances close to the data manifold.”
Counterfactual Explanations and Algorithmic Recourses for Machine Learning: A Review (2020, neutral): “Adherence to the data manifold has been addressed using several different approaches, like training VAEs on the data distribution [78, 159, 210, 311], constraining the distance of a counterfactual from the k nearest training datapoints [67, 89, 164], directly sampling points from the latent space of”
Explainable AI (XAI): Core Ideas, Techniques, and Solutions (2022, background): “The two categories of explanations includes Pertinent Positives and Pertinent Negatives [12].”
A Survey on Neural Network Interpretability (2020, background): “xn are absent that x is classified as ŷ” [38].”
Questioning the AI: Informing Design Practices for Explainable AI User Experiences (2020, methodology): “perturbed, absent or present [27, 91, 100] Why, Why not, How to be that”
A Primer on Zeroth-Order Optimization in Signal Processing and Machine Learning: Principals, Recent Advances, and Applications (2020, methodology): “Here, we apply ZO optimization to generating contrastive explanations [76] for two ML applications – handwritten”
The hidden assumptions behind counterfactual explanations and principal reasons (2019, background): “Recent proposals from computer scientists have focused on generating counterfactual explanations for the decisions of a machine learning model [27, 47, 44, 29, 20, 17, 15, 34, 25, 9].”
Teach Me to Explain: A Review of Datasets for Explainable Natural Language Processing (2021, background): “[31] Amit Dhurandhar, Pin-Yu Chen, Ronny Luss, Chun-Chen Tu, Paishun Ting, Karthikeyan Shanmugam, and Payel Das.”
Towards Realistic Individual Recourse and Actionable Explanations in Black-Box Decision Making Systems (2019, background): “Note that we do not view the proposed mechanism as a means to mitigate liability issues for the institution deploying algorithmic decision–making, which is the focus of providing counterfactual explanations (Wachter et al., 2017; Dhurandhar et al., 2018).”
FOCUS: Flexible Optimizable Counterfactual Explanations for Tree Ensembles (2019, methodology): “Counterfactual examples have been used in a variety of ML areas, such as reinforcement learning [23], deep learning [1], and explainable AI (XAI) [8, 13, 17, 18, 22, 24, 26, 28, 31, 35, 36].”
Equalizing Recourse across Groups (2019, methodology): “[DCL + 18] recently proposed a method to provide minimal absent feature sets from examples which are sufﬁcient for ﬂipping their prediction.”
Understanding Uncertainty in Bayesian Neural Networks (2019, background): “The most common form of machine learning model explanation is the saliency map (Chang et al., 2019; Dabkowski and Gal, 2017; Dhurandhar et al., 2018; Lundberg and Lee, 2017; Shrikumar et al., 2017; Sundararajan et al., 2017; Zintgraf et al., 2017).”
Proactive Pseudo-Intervention: Causally Informed Contrastive Learning For Interpretable Vision Models (2020, background): “Contrastive explanations are produced in [13] to justify the predictions from a deep neural network.”
Generating Counterfactual Explanations with Natural Language (2018, background): “Copyright 2018 by the author(s). some input belongs to one class and not another (Wachter et al., 2017; Dhurandhar et al., 2018).”