whose relevant path is followed and used as an explanation

[suchona.kani.z]

On the left we have data such as images, numerical or other types of data. These go through a model, here called a black box model because it is not clear what is happening inside. After that, a result is output, such as that the image is a cat with a probability of 99 percent or a prediction for a dividing line for numerical data. There are three ways to apply Explainable AI in such a construct:

Model Explanation: This is about explaining the model and not just a single decision made by the model. One way to do this is to use an interpretable machine learning model that is trained on the output of the black box using the input of the test variables. This allows you to simulate the black box and by using a white box model you can understand and explain the results. Model Explanation deals with the model and the results produced by the black box model.

Model Inspection: This involves creating a representation to explain armenia consumer email list​ the properties of the black box model or predictions. This can be done, for example, by using the output of test instances to generate a graph. For example, sensitivity to variable changes would be one way to represent the model as a graph.

Outcome Explanation: This procedure involves providing an explanation for a single input. It is not about understanding the entire model, but only a justification for a decision. An example of this could be a decision tree,

In general, Explainable AI is relatively easy to implement for machine learning methods, as there are a relatively small number of components that make a decision. With deep learning and the neural networks used in it, this is much more difficult, as the number of components can easily run into the thousands or millions. One of the most common networks for natural language processing - GPT-3 - has a number of 175 billion parameters - this makes Explainable AI significantly more difficult, but not impossible.

Examples of Explainable AI
I would like to show you a few examples of what Explainable AI can look like and how it can be used to explain or understand decisions. There is no detailed standardized procedure like we use for our own applications, but it is intended to give a brief insight into simple possibilities. I would like to show these procedures in more detail with examples in another blog article.