Google Play icon

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

Share
Posted May 24, 2019
This news or article is intended for readers with certain scientific or professional knowledge in the field.

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

Click to open (PDF)

This short paper introduces an approach to producing explanations or justifications of decisions made in some artificial intelligence and machine learning (AI/ML) systems, using methods derived from those for fault location in combinatorial testing. NIST specialists show that validation and explainability issues are closely related to the problem of fault location in combinatorial testing, and that certain methods and tools developed for fault location can also be applied to this problem.

This approach is particularly useful in classification problems, where the goal is to determine an object’s membership in a set based on its characteristics. We use a conceptually simple scheme to make it easy to justify classification decisions: identifying combinations of features that are present in members of the identified class but absent or rare in non-members. The method has been implemented in a prototype tool called ComXAI, and examples of its application are given. Examples from a range of application domains are included to show the utility of these methods.

The public comment period for this document ends on July 3, 2019.  See the document details for a copy of the paper and instructions for submitting comments.

Abstract

This short paper introduces an approach to producing explanations or justifications of decisions made in some artificial intelligence and machine learning (AI/ML) systems, using methods derived from those for fault location in combinatorial testing. We show that validation and explainability issues are closely related to the problem of fault location in combinatorial testing, and that certain methods and tools developed for fault location can also be applied to this problem. This approach is particularly useful in classification problems, where the goal is to determine an object’s membership in a set based on its characteristics. We use a conceptually simple scheme to make it easy to justify classification decisions: identifying combinations of features that are present in members of the identified class but absent or rare in non-members. The method has been implemented in a prototype tool called ComXAI, and examples of its application are given. Examples from a range of application domains are included to show the utility of these methods.

Keywords

artificial intelligence (AI); assurance of autonomous systems; combinatorial testing; covering array; explainable AI; machine learning

Source: NIST

Featured news from related categories:

Technology Org App
Google Play icon
83,394 science & technology articles

Most Popular Articles

  1. Bright Fireball Explodes Over Ontario, Meteorite Fragments Might Have Reached the Ground (August 5, 2019)
  2. Why older people smell the way they do? Japanese have even a special word for it (August 4, 2019)
  3. Terraforming the Surface of Mars with Silica Aerogel? (July 23, 2019)
  4. Moisturizers May Be Turning Your Skin Into ‘Swiss Cheese’ (4 days old)
  5. Swarm Autonomy Tested in Second Major DARPA OFFSET Field Experiment (August 8, 2019)

Follow us

Facebook   Twitter   Pinterest   Tumblr   RSS   Newsletter via Email