aiqbench analysis
This paper introduces CLIQUE, a novel model-agnostic method for calculating local variable importance that addresses limitations in existing tools like LIME and SHAP regarding multi-class classification and loss space characterization. By leveraging quantile expectations to capture non-linear interactions, the technique offers improved stability over permutation-based approaches while correctly assigning zero importance where model responses are invariant. This advancement provides researchers with a more robust framework for interpreting complex machine learning models across diverse datasets.
Conditional Local Importance by Quantile Expectations
arXiv:2411.08821v4 Announce Type: replace-cross Abstract: Global variable importance measures are commonly used to interpret the results of machine learning models. Local variable importance techniques assess how variables contribute to individual observations. Current, popular methods, including LIME and SHAP, provide useful measures of feature contribution in the prediction space, while leaving opportunities for improved characterization of local structure in the model loss space. Additionally, they are not natively adapted for multi-class classification problems. We propose a new model-agnostic method for calculating local variable importance, CLIQUE, that highlights locally dependent relationships, provides improved stability over permutation-based methods, and can be directly applied to multi-class classification problems. Simulated and real-world examples show that CLIQUE emphasizes locally dependent information, captures interaction behavior beyond what can be evaluated by correlations, and assigns zero importance in regions where the response is invariant to changes in variables.
Abstract:Global variable importance measures are commonly used to interpret the results of machine learning models. Local variable importance techniques assess how variables contribute to individual observations. Current, popular methods, including LIME and SHAP, provide useful measures of feature contribution in the prediction space, while leaving opportunities for improved characterization of local structure in the model loss space. Additionally, they are not natively adapted for multi-class classification problems. We propose a new model-agnostic method for calculating local variable importance, CLIQUE, that highlights locally dependent relationships, provides improved stability over permutation-based methods, and can be directly applied to multi-class classification problems. Simulated and real-world examples show that CLIQUE emphasizes locally dependent information, captures interaction behavior beyond what can be evaluated by correlations, and assigns zero importance in regions where the response is invariant to changes in variables.
Machine Learning (stat.ML); Machine Learning (cs.LG); Computation (stat.CO)
arXiv:2411.08821 [stat.ML]
(or arXiv:2411.08821v4 [stat.ML] for this version)
Full text under the publisher's open licence · 151 words
Related coverage
The Entangled Dimensions of Decision Theory
LessWrong · Jul 30, 2026
Verifiable Regularity Criterion for Conditional Expectation Operators and Conditional Mean Embeddings with Applications to Nonparametric Regression, Bayesian Inverse Problems, and Koopman Operators
arXiv · Aug 7, 2026
Quantum Sufficiency for Self-Adjoint Statistical Models via Likelihood-Type Operators on Real $*$-Subalgebras and Real Jordan Algebras
arXiv · Sep 23, 2026
Generalized Kalman filter based temporal difference reinforcement learning
arXiv · Jul 23, 2026