Thenn2poly package implements the NN2Poly methodthat allows to transform an already trained deep feed-forward fullyconnected neural network into a polynomial representation that predictsas similar as possible to the original neural network. The obtainedpolynomial coefficients can be used to explain features (and theirinteractions) importance in the neural network, therefore working as atool for interpretability or eXplainable Artificial Intelligence(XAI).
Pablo Morala, J. Alexandra Cifuentes, Rosa E. Lillo, Iñaki Ucar(2021). “Towards a mathematical framework to inform neural networkmodelling via polynomial regression.”Neural Networks,142, 57-72. doi:10.1016/j.neunet.2021.04.036
Pablo Morala, J. Alexandra Cifuentes, Rosa E. Lillo, Iñaki Ucar(2023). “NN2Poly: A Polynomial Representation for Deep Feed-ForwardArtificial Neural Networks.”IEEE Transactions on Neural Networksand Learning Systems, (Early Access). doi:10.1109/TNNLS.2023.3330328
The latest release version available in CRAN can be installed as:
install.packages("nn2poly")The installation from GitHub requires theremotespackage.
# install.packages("remotes")remotes::install_github("IBiDat/nn2poly")This package is part of the project/grant PDC2022-133359-I00 fundedby MCIN/AEI/10.13039/501100011033 and by the European Union“NextGenerationEU/PRTR”.
