Explaining Classifiers with Causal Concept Effect (CaCE)

Abstract

How can we understand classification decisions made by deep neural networks? Many existing explainability methods rely solely on correlations and fail to account for confounding, which may result in potentially misleading explanations. To overcome this problem, we define the Causal Concept Effect (CaCE) as the causal effect of (the presence or absence of) a human-interpretable concept on a deep neural net's predictions. We show that the CaCE measure can avoid errors stemming from confounding. Estimating CaCE is difficult in situations where we cannot easily simulate the do-operator. To mitigate this problem, we use a generative model, specifically a Variational AutoEncoder (VAE), to measure VAE-CaCE. In an extensive experimental analysis, we show that the VAE-CaCE is able to estimate the true concept causal effect, compared to baselines for a number of datasets including high dimensional images.

References in corpus (8)

• Towards A Rigorous Science of Interpretable Machine Learning
• SmoothGrad: removing noise by adding noise
• Model-Agnostic Interpretability of Machine Learning
• Benchmarking Attribution Methods with Relative Feature Importance
• Explaining Image Classifiers by Counterfactual Generation
• CausalGAN: Learning Causal Implicit Generative Models with Adversarial Training
• Generating Counterfactual Explanations with Natural Language
• Methods and Models for Interpretable Linear Classification

Cited by in corpus (30)

• Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey
• Post-hoc Interpretability for Neural NLP: A Survey
• Acquisition of Chess Knowledge in AlphaZero
• Entropy-based Logic Explanations of Neural Networks
• Causality Learning: A New Perspective for Interpretable Machine Learning
• Model-Based Counterfactual Synthesizer for Interpretation
• Counterfactual Generative Networks
• Now You See Me (CME): Concept-based Model Extraction
• Instance-wise or Class-wise? A Tale of Neighbor Shapley for Concept-based Explanation
• Generative causal explanations of black-box classifiers
• Best of both worlds: local and global explanations with human-understandable concepts
• Causal Interpretability for Machine Learning -- Problems, Methods and Evaluation
• CARLA: A Python Library to Benchmark Algorithmic Recourse and Counterfactual Explanation Algorithms
• Explaining in Style: Training a GAN to explain a classifier in StyleSpace
• Debiasing Concept-based Explanations with Causal Analysis
• ELLA: Exploration through Learned Language Abstraction
• Amnesic Probing: Behavioral Explanation with Amnesic Counterfactuals
• Learning Deep Attribution Priors Based On Prior Knowledge
• Statistically Significant Concept-based Explanation of Image Classifiers via Model Knockoffs
• DISSECT: Disentangled Simultaneous Explanations via Concept Traversals
• Leveraging Sparse Linear Layers for Debuggable Deep Networks
• Born Identity Network: Multi-way Counterfactual Map Generation to Explain a Classifier's Decision
• Cause and Effect: Hierarchical Concept-based Explanation of Neural Networks
• MEME: Generating RNN Model Explanations via Model Extraction
• Unifying Model Explainability and Robustness via Machine-Checkable Concepts
• NeuroView: Explainable Deep Network Decision Making
• What if This Modified That? Syntactic Interventions via Counterfactual Embeddings
• Matched sample selection with GANs for mitigating attribute confounding
• A Glitch in the Matrix? Locating and Detecting Language Model Grounding with Fakepedia
• Knowledge graphs for empirical concept retrieval