Interpreting Neural Network Judgments via Minimal, Stable, and Symbolic Corrections

Abstract

We present a new algorithm to generate minimal, stable, and symbolic corrections to an input that will cause a neural network with ReLU activations to change its output. We argue that such a correction is a useful way to provide feedback to a user when the network's output is different from a desired output. Our algorithm generates such a correction by solving a series of linear constraint satisfaction problems. The technique is evaluated on three neural network models: one predicting whether an applicant will pay a mortgage, one predicting whether a first-order theorem can be proved efficiently by a solver using certain heuristics, and the final one judging whether a drawing is an accurate rendition of a canonical drawing of a cat.

arXiv comment24 pages

Cited by in corpus (17)

• Explainability Techniques for Graph Convolutional Networks
• A survey of algorithmic recourse: definitions, formulations, solutions, and prospects
• MoËT: Mixture of Expert Trees and its Application to Verifiable Reinforcement Learning
• Generative causal explanations of black-box classifiers
• Explaining Explanations to Society
• Interpretable Machine Learning: Moving From Mythos to Diagnostics
• Learning outside the Black-Box: The pursuit of interpretable models
• Neuro-Symbolic Artificial Intelligence: Current Trends
• Synthesizing Action Sequences for Modifying Model Decisions
• A Zero-Positive Learning Approach for Diagnosing Software Performance Regressions
• Robust and Stable Black Box Explanations
• Explaining Groups of Points in Low-Dimensional Representations
• GRACE: Generating Concise and Informative Contrastive Sample to Explain Neural Network Model's Prediction
• Fair Normalizing Flows
• Scaling Symbolic Methods using Gradients for Neural Model Explanation
• Learning Models for Actionable Recourse
• Automated Dependence Plots