Ranking importance of input parameters of neural networks

By Andrew H. Sung. 1998

How researchers see this paper

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

Citations in the corpus

Methods for interpreting and understanding deep neural networks (2017, background): “A first approach to identify the most important input features is sensitivity analysis [77,66,29].”
Drug discovery with explainable artificial intelligence (2020, methodology): “Alongside the well-established step-wise approaches, 83–85 methods such as feature masking, 86 perturbation analysis, 87 response randomization, 88 and conditional multivariate models 89 belong to this category.”
Review and comparison of methods to study the contribution of variables in artificial neural network models (2003, background): “the most relevant to the construction of a network with a small MSE ( Sung, 1998 ).”
Bond Risk Premia with Machine Learning∗ (2018, unknown): “For instance, Sung (1998) proposed a stepwise method that consists of adding or rejecting one input variable at a time and noting the effect on the Mean Squared Error (MSE).”
Opening the black box - data driven visualization of neural networks (2005, methodology): “The stepwise methods [20, 27] add or reject one input variable at a time and the MSE (mean square error) of the output is used to identify the most important variables.”
Neural networks for classification: a survey (2000, methodology): “Sung [162] studies three methods of sensitivity analysis, fuzzy curves, and change of mean square error to rank input feature importance.”
Explainability Techniques for Graph Convolutional Networks (2019, background): “At high level these techniques can be divided into those that attempt to interpret the model as a whole (Simonyan et al., 2014; Nguyen et al., 2016) and those that try to explain individual predictions made by a model (Sung, 1998; Bach et al., 2015) 1 .”
Identifying important features for intrusion detection using support vector machines and neural networks (2003, methodology): “Once the data is classified into two classes, a suitable optimizing algorithm can be used if necessary for further feature identification, depending on the application [14].”