Back to Search
Start Over
Interpreting Deep Learning Models with Marginal Attribution by Conditioning on Quantiles
- Source :
- Data Mining and Knowledge Discovery, 36 (4)
- Publication Year :
- 2021
- Publisher :
- arXiv, 2021.
-
Abstract
- A vast and growing literature on explaining deep learning models has emerged. This paper contributes to that literature by introducing a global gradient-based model-agnostic method, which we call Marginal Attribution by Conditioning on Quantiles (MACQ). Our approach is based on analyzing the marginal attribution of predictions (outputs) to individual features (inputs). Specifically, we consider variable importance by fixing (global) output levels, and explaining how features marginally contribute to these fixed global output levels. MACQ can be seen as a marginal attribution counterpart to approaches such as accumulated local effects, which study the sensitivities of outputs by perturbing inputs. Furthermore, MACQ allows us to separate marginal attribution of individual features from interaction effects and to visualize the 3-way relationship between marginal attribution, output level, and feature value.<br />Data Mining and Knowledge Discovery, 36 (4)<br />ISSN:1384-5810<br />ISSN:1573-756X
- Subjects :
- QA75
FOS: Computer and information sciences
Computer Science - Machine Learning
Variable importance
Interaction
Computer Networks and Communications
HB
Machine Learning (stat.ML)
Space (commercial competition)
Statistics - Applications
Accumulated local effects (ALE)
Machine Learning (cs.LG)
FOS: Economics and business
Attribution
Post-hoc analysis
Statistics - Machine Learning
Feature (machine learning)
Econometrics
Applications (stat.AP)
Mathematics
Partial dependence plot (PDP)
Statistical Finance (q-fin.ST)
T1
business.industry
Deep learning
Quantitative Finance - Statistical Finance
Explainable AI (XAI)
Computer Science Applications
Variable (computer science)
Model-agnostic tools
68T07
Conditioning
Artificial intelligence
business
Locally interpretable model-agnostic explanation (LIME)
Value (mathematics)
Quantile
Information Systems
Subjects
Details
- ISSN :
- 13845810 and 1573756X
- Database :
- OpenAIRE
- Journal :
- Data Mining and Knowledge Discovery, 36 (4)
- Accession number :
- edsair.doi.dedup.....cd1f8df526d964780405b85cd1645277
- Full Text :
- https://doi.org/10.48550/arxiv.2103.11706