Yuya Yoshikawa, Tomoharu Iwata, Explanation-Based Training with Differentiable Insertion/Deletion Metric-Aware Regularizers, The 27th International Conference on Artificial Intelligence and Statistics (AISTATS2024), Valencia, Spain, May 2024.

Package idexpo can be installed in this repository by the following command:

pip install .

First, please download ResNet-18 pre-trained weights trained on CIFAR-10 (resnet18_cifar10.pth) from Google Drive, and place it into examples/weights directory. Then, you can fine-tune ResNet-18 with ID-ExpO and evaluate the fine-tuned model.

cd examples
python train_cifar10_gradcam.py

Please see LICENSE.txt.

• Add implementation for Grad-CAM
• Add implementation for LIME
• Add implementation for tabular data