Review and comparison of methods to study the contribution of variables in artificial neural network models

By M. Gevrey, I. Dimopoulos, Sovan Lek. 2003

1,171 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Environmental Engineering, Nature and Landscape Conservation, Evaluation & Benchmarks, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Environmental Engineering

In Nature and Landscape Conservation

In Evaluation & Benchmarks

In Safety, Robustness & Interpretability

Citations in the corpus

Methods for interpreting and understanding deep neural networks (2017, methodology): “Sensitivity analysis has been regularly used in scientific applications of machine learning such as medical diagnosis [25], ecological modeling [18], or mutagenicity prediction [6].”
Explaining Recurrent Neural Network Predictions in Sentiment Analysis (2017, background): “We refer to the above definition of relevance as Sensitivity Analysis (SA) (Dimopoulos et al., 1995; Gevrey et al., 2003).”
Explaining nonlinear classification decisions with deep Taylor decomposition (2015, methodology): “Figure 1 illustrates on a simple two-dimensional function the difference between this decomposition technique and another frequently used explanation technique called sensitivity analysis [44], which explains the prediction as”
On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation (2015, methodology): “Quantifying the importance of input variables using a neural network model has also been studied in specific areas such as ecological modeling, where [29, 30] surveyed a large ensemble of possible analyses, including, computing partial derivatives, perturbation analysis, weights analysis, and studyi”
All Models are Wrong, but Many are Useful: Learning a Variable’s Importance by Studying an Entire Class of Prediction Models Simultaneously (2018, methodology): “We focus on the empirical versions of our importance metrics (M̂R, M̂CR− and M̂CR+), and compare them against AR, which is typically interpreted as an in-sample measure (Breiman, 2001), or as an intermediate step to estimate an alternate importance measure in terms of variable rankings (Gevrey et al”
"What is relevant in a text document?": An interpretable machine learning approach (2016, methodology): “Sensitivity analysis (SA) [20, 24, 25] assigns scores Ri,t = (∂xk/∂xi,t) 2 to input variables representing the steepness of the decision function in the input space.”
Explaining Predictions of Non-Linear Classifiers in NLP (2016, background): “An alternative procedure called sensitivity analysis (SA) produces explanations by scoring input variables based on how they affect the decision output locally (Dimopoulos et al., 1995; Gevrey et al., 2003).”
Understanding Patch-Based Learning by Explaining Predictions (2018, background): “Additional explanations are given for the same 3-dimensional convolutional neural network architecture untrained as well as using gradient-based sensitivity analysis [4, 23] for comparison.”
Explainable Artificial Intelligence: a Systematic Review (2020, methodology): “Lastly, [333] compared the completeness of the explanations generated by seven methods for explainability that interpret the logic followed by DNNs by calculating the partial derivatives of the output according to the input variables, analysing the network’s weights or perturbing the input variables”
Explaining the unique nature of individual gait patterns with deep learning (2018, background): “The lack of understanding and interpreting the decision process of machine learning techniques is a clear drawback and recently attracted attention in the field of machine learning [4, 44, 22, 59, 77, 3, 81, 48, 79, 58, 20, 43].”