Fundamental of Explainable AI (XAI)

The modern AI is based on deep neural network (DNN) models. These DNNs have found to be highly effective in across various tasks ranging from language generation, and understanding to computer vision. However, these models are considered as Blackbox in which, we do not know, how it process input data and how it generates the decision (predictions). This limitation have led to implication for safety and responsibility. In such regard, XAI are state of art technique that helps us to open such Blackbox of DNN models and understand (a) way model process data (b) what feature of data impact model decision (c) which layer of DNN play a significant role in model decision (d) contrafactual test which help to assess correlation in between input and output decision.