Use of Visual Analytics (VA) in Explainable Artificial Intelligence (XAI): A Framework of Information Granules

Research work has been increasing in Visual Analytics (VA) since it was first defined in 2005. The techniques of VA that integrated Machine Learning (ML) models and interactive visualizations have formed a human-centered machine learning approach to assist in Data Analytics. VA aims to interpret the complexities of Big Data and underlying ML models by engaging analysts in an iterative process of observing, interpreting, and evaluating inputs, outputs, and architectures of these models. The process then subsequently provides guidance to users, interaction techniques to control AI, and information about inner workings that are often hidden. This chapter defines underlying stages of ML pipeline in feature selection and model performance as components of Information Granules (IG) in VA; it also explores the use of VA in XAI. This study reviews 13 top-tier publications in recent VA literature to demonstrate (1) the interoperability strategies of VA in feature relevance, model performance, and model architecture; (2) global and local interpretability in information and visual scalability; and (3) stability of explanations through user case studies, reusability, and design goals. The chapter also analyzes the current stage of VA for granular computing in the end. The future work of VA scientists will be to focus on broader behaviors of ML models, particularly in Neural Network, to gain public trust for AI.