Share

Combinatorial Methods for Explainability in AI and Machine Learning – Draft White Paper

A breeze white paper is now accessible for comment, An Application of Combinatorial Methods for Explainability in Artificial Intelligence and Machine Learning.

Click to open (PDF)

This brief paper introduces an proceed to producing explanations or justifications of decisions done in some synthetic comprehension and appurtenance training (AI/ML) systems, regulating methods subsequent from those for error plcae in combinatorial testing. We uncover that validation and explainability issues are closely associated to a problem of error plcae in combinatorial testing, and that certain methods and collection grown for error plcae can also be practical to this problem. This proceed is quite useful in sequence problems, where a idea is to establish an object’s membership in a set formed on a characteristics. We use a conceptually elementary intrigue to make it easy to clear sequence decisions: identifying combinations of facilities that are benefaction in members of a identified category though absent or singular in non-members. The process has been implemented in a antecedent apparatus called ComXAI, and examples of a focus are given. Examples from a operation of focus domains are enclosed to uncover a focus of these methods.

The open criticism period for this request ends on Jul 3, 2019.  See a request sum for a duplicate of a paper and instructions for submitting comments.

Abstract

This brief paper introduces an proceed to producing explanations or justifications of decisions done in some synthetic comprehension and appurtenance training (AI/ML) systems, regulating methods subsequent from those for error plcae in combinatorial testing. We uncover that validation and explainability issues are closely associated to a problem of error plcae in combinatorial testing, and that certain methods and collection grown for error plcae can also be practical to this problem. This proceed is quite useful in sequence problems, where a idea is to establish an object’s membership in a set formed on a characteristics. We use a conceptually elementary intrigue to make it easy to clear sequence decisions: identifying combinations of facilities that are benefaction in members of a identified category though absent or singular in non-members. The process has been implemented in a antecedent apparatus called ComXAI, and examples of a focus are given. Examples from a operation of focus domains are enclosed to uncover a focus of these methods.

Keywords

artificial comprehension (AI); declaration of unconstrained systems; combinatorial testing; covering array; explainable AI; appurtenance learning

Source: NIST


<!–

Comment this news or article

–>