Explainable Artificial Intelligence
(Autumn 2021)

Hidden neuron analysis methods:

• Visualize, revert-map or label features that are learned by hidden neurons

• GAN

• Maximum Activation

• Qualitatively analyzing every neuron does not provide much actionable and quantitative interpretation about the overall mechanism of the whole model

Model mimicking methods: imitate the classification function of a target model and build a transparent model that is easy to interpret and has high accuracy

• Due to the reduced model complexity of a mimic model, no guarantee that a deep model with a large VC dimension can be successfully imitated by a simpler shallow model

Local interpretation methods:

• Compute and visualize the important features for an input instance by analyzing the predictions of its local perturbations

• LIME

• Saliency map

• Perceptively indistinguishable instances may not be explained consistently