Table 1 Classification of XAI Techniques

From: Explainable AI: from black box to glass box

GlobalEnforce interpretability constraints into the structure and learning mechanisms of deep learning modelsDevelop interpretable global surrogate models based on input-output associations predicted by a black-box model
Apply diagnostic techniques to understand the importance of specific features in a black-box model’s predictions
LocalUse attention mechanisms to show how the model selectively focuses on features in high-dimensional input for an instanceDevelop interpretable surrogate models with local fidelity in the vicinity of an instance