Explain the model decision

Even if we cannot fully understand the decisions or results of LLMs, we can understand certain aspects that led to this decision. But first, we need to answer and understand what we are trying to explain here. We have three aspects for it as shown in the following taxonomy diagram:

Figure 11.16 – Taxonomy of XAI in 3D

First, are we trying to understand either how a model makes decisions as a whole or its behavior in a specific input? This distinction is referred to as the global explanation and local explanation. While local explainers deal with the decision on a specific input, global explainers try to explain the model’s prediction process as a whole.

The second distinction between XAI models is whether they are self-explaining or post-hoc models. In the former approach, also known as self-explainers, models make both predictions and explanations as seen in self-attention mechanisms. However, the self-attention...