R implementation of Contextual Importance and Utility for Explainable AI with Images.
CIU was developed by Kary Främling in his PhD thesis Learning and Explaining Preferences with Neural Networks for Multiple Criteria Decision Making, (written in French, title Modélisation et apprentissage des préférences par réseaux de neurones pour l'aide à la décision multicritère), available online for instance here: https://tel.archives-ouvertes.fr/tel-00825854/document. It was originally implemented in Matlab and has later been re-implemented in Python and R (package ciu) for tabular data.
This ciu.image package implements CIU for image recognition "explanation" using e.g. saliency maps.
In the future, ciu.image will presumably be available from CRAN. Meanwhile, it can be installed directly from Github with the command
# install.packages('devtools') # Uncomment if devtools wasn't installed already
devtools::install_github('KaryFramling/ciu.image')
The root directory contains several source files with example code for using ciu.image. A simple example using a kitten image with VGG16 is shown here.
library(keras)
library(lime)
library(magick)
library(ciu.image)
vgg_predict_function <- function(model, imgpath) {
predict(model, image_prep(imgpath))
}
image_prep <- function(x) {
arrays <- lapply(x, function(path) {
img <- image_load(path, target_size = c(224,224))
x <- image_to_array(img)
x <- array_reshape(x, c(1, dim(x)))
x <- imagenet_preprocess_input(x)
})
do.call(abind::abind, c(arrays, list(along = 1)))
}
model <- application_vgg16(
weights = "imagenet",
include_top = TRUE
)
model_labels <- readRDS(system.file('extdata', 'imagenet_labels.rds', package='ciu.image'))
imgpath <- system.file('extdata', 'kitten.jpg', package = 'ciu.image')
ciu <- ciu.image.new(model, vgg_predict_function, output.names = model_labels)
plist <- ciu$plot.image.explanation(imgpath, ind.output = c(1,2,3))
for ( i in 1:length(plist) )
print(plist[[i]])
The first publication on CIU was in the ICANN conference in Paris in 1995: FRÄMLING, Kary, GRAILLOT, Didier. Extracting Explanations from Neural Networks. ICANN'95 proceedings, Vol. 1, Paris, France, 9-13 October, 1995. Paris: EC2 & Cie, 1995. pp. 163-168., accessible at http://www.cs.hut.fi/u/framling/Publications/FramlingIcann95.pdf.
The second publication, and last before "hibernation" of CIU research, is FRÄMLING, Kary. Explaining Results of Neural Networks by Contextual Importance and Utility. Proceedings of the AISB'96 conference, 1-2 April 1996. Brighton, UK, 1996., accessible at http://www.cs.hut.fi/u/framling/Publications/FramlingAisb96.pdf.
The first publication after "hibernation" is ANJOMSHOAE, Sule, FRÄMLING, Kary, NAJJAR, Amro. Explanations of black-box model predictions by contextual importance and utility. In: Lecture Notes in Computer Science, Vol. 11763 LNAI; Revised Selected Papers of Explainable, Transparent Autonomous Agents and Multi-Agent Systems - 1st International Workshop, EXTRAAMAS 2019, Montreal, Canada, May 13-14, 2019. pp. 95-109. https://www.researchgate.net/publication/333310978_Explanations_of_Black-Box_Model_Predictions_by_Contextual_Importance_and_Utility.