Machine Learning has seen widespread industry adoption in the last couple of years. One of the breakthrough adoption in machine learning is ability to explain or model interpretation which as a concept is still theoretical and subjective. There are still no clear boundry to decide which part of the AI model should be explained, and what should be the criteria to decide which part of the AI model should explained?
- should we explain the feature/s, what the extracted feature looks like?
- should we explain why ceirtain key features are selected for decision and other not?
- should we explain the model output like how the model take decision?
It could be important to explain all of them; but, do we have any boundry conditions to decide which to explain and to whom?
In order to perform our imaginative work/tasks through a machine in most faster way:
1) We need to know about the structure and kind of the data, e.g., sound/voice data, still images data, moving images/objects data (videos) etc.
2) In the second step, it is important to understand which models (functions or functional, that are input to output general equations, preferably analytic with closed form)...
3) In the 3rd step it is important study which model (from many available model) is super optimal for the given data.
4) Model/Function or mathematical/statistical data operator or processor having optimal properties of accuracy on the given data for a particular machine learning task is a model selection process.....
5) At this step train the model and generalize via new unseen data for decisions is nothing but AI...
.
the following paper makes a great job at clarifying such issues :
https://www.researchgate.net/publication/338184751
there is no single answer to your question and you cannot progress as long as you do not specify the audience which should benefit from your explainations
.
It is important to firstly know who you are going to explain the model to. But generally, it could be important to know how the model makes the decision and why it made it. Furthermore, it could also be helpful to know the correlation and the importance of the features in relation to the decision made.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence