Objective criteria for explanations of machine learning models
Abstract Objective criteria to evaluate the performance of machine learning (ML) model explanations are a critical ingredient in bringing greater rigor to the field of explainable artificial intelligence. In this article, we survey three of our proposed criteria that each target different classes of...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2021-12-01
|
| Series: | Applied AI Letters |
| Subjects: | |
| Online Access: | https://doi.org/10.1002/ail2.57 |
| _version_ | 1818905490290114560 |
|---|---|
| author | Chih‐Kuan Yeh Pradeep Ravikumar |
| author_facet | Chih‐Kuan Yeh Pradeep Ravikumar |
| author_sort | Chih‐Kuan Yeh |
| collection | DOAJ |
| description | Abstract Objective criteria to evaluate the performance of machine learning (ML) model explanations are a critical ingredient in bringing greater rigor to the field of explainable artificial intelligence. In this article, we survey three of our proposed criteria that each target different classes of explanations. In the first, targeted at real‐valued feature importance explanations, we define a class of “infidelity” measures that capture how well the explanations match the ML models. We show that instances of such infidelity minimizing explanations correspond to many popular recently proposed explanations and, moreover, can be shown to satisfy well‐known game‐theoretic axiomatic properties. In the second, targeted to feature set explanations, we define a robustness analysis‐based criterion and show that deriving explainable feature sets based on the robustness criterion yields more qualitatively impressive explanations. Lastly, for sample explanations, we provide a decomposition‐based criterion that allows us to provide very scalable and compelling classes of sample‐based explanations. |
| first_indexed | 2024-12-19T21:24:10Z |
| format | Article |
| id | doaj.art-0c6d23f22d4c4605901f9151d4c259b9 |
| institution | Directory Open Access Journal |
| issn | 2689-5595 |
| language | English |
| last_indexed | 2024-12-19T21:24:10Z |
| publishDate | 2021-12-01 |
| publisher | Wiley |
| record_format | Article |
| series | Applied AI Letters |
| spelling | doaj.art-0c6d23f22d4c4605901f9151d4c259b92022-12-21T20:05:10ZengWileyApplied AI Letters2689-55952021-12-0124n/an/a10.1002/ail2.57Objective criteria for explanations of machine learning modelsChih‐Kuan Yeh0Pradeep Ravikumar1Machine Learning Department, Carnegie Mellon University Pittsburgh Pennsylvania USAMachine Learning Department, Carnegie Mellon University Pittsburgh Pennsylvania USAAbstract Objective criteria to evaluate the performance of machine learning (ML) model explanations are a critical ingredient in bringing greater rigor to the field of explainable artificial intelligence. In this article, we survey three of our proposed criteria that each target different classes of explanations. In the first, targeted at real‐valued feature importance explanations, we define a class of “infidelity” measures that capture how well the explanations match the ML models. We show that instances of such infidelity minimizing explanations correspond to many popular recently proposed explanations and, moreover, can be shown to satisfy well‐known game‐theoretic axiomatic properties. In the second, targeted to feature set explanations, we define a robustness analysis‐based criterion and show that deriving explainable feature sets based on the robustness criterion yields more qualitatively impressive explanations. Lastly, for sample explanations, we provide a decomposition‐based criterion that allows us to provide very scalable and compelling classes of sample‐based explanations.https://doi.org/10.1002/ail2.57explainable AIfeature importanceimportant feature setsobjective evaluation criteriasample explanations |
| spellingShingle | Chih‐Kuan Yeh Pradeep Ravikumar Objective criteria for explanations of machine learning models Applied AI Letters explainable AI feature importance important feature sets objective evaluation criteria sample explanations |
| title | Objective criteria for explanations of machine learning models |
| title_full | Objective criteria for explanations of machine learning models |
| title_fullStr | Objective criteria for explanations of machine learning models |
| title_full_unstemmed | Objective criteria for explanations of machine learning models |
| title_short | Objective criteria for explanations of machine learning models |
| title_sort | objective criteria for explanations of machine learning models |
| topic | explainable AI feature importance important feature sets objective evaluation criteria sample explanations |
| url | https://doi.org/10.1002/ail2.57 |
| work_keys_str_mv | AT chihkuanyeh objectivecriteriaforexplanationsofmachinelearningmodels AT pradeepravikumar objectivecriteriaforexplanationsofmachinelearningmodels |