How does this interaction affect me? Interpretable attribution for feature interactions

TL;DR

This paper introduces Archipelago, a scalable and interpretable framework for attributing and detecting feature interactions in machine learning models, enhancing transparency and understanding of complex predictions.

Contribution

The paper presents a novel interaction attribution framework called Archipelago that improves interpretability and scalability in analyzing feature interactions in prediction models.

Findings

Archipelago provides significantly more interpretable explanations than existing methods.

The approach offers new insights into deep neural network behavior.

Experiments demonstrate effectiveness on standard annotation labels.

Abstract

Machine learning transparency calls for interpretable explanations of how inputs relate to predictions. Feature attribution is a way to analyze the impact of features on predictions. Feature interactions are the contextual dependence between features that jointly impact predictions. There are a number of methods that extract feature interactions in prediction models; however, the methods that assign attributions to interactions are either uninterpretable, model-specific, or non-axiomatic. We propose an interaction attribution and detection framework called Archipelago which addresses these problems and is also scalable in real-world settings. Our experiments on standard annotation labels indicate our approach provides significantly more interpretable explanations than comparable methods, which is important for analyzing the impact of interactions on predictions. We also provide accompanying visualizations of our approach that give new insights into deep neural networks.