EXplainable Artificial Intelligence (XAI)

EXplainable Artificial Intelligence (XAI)

This repository includes the codes for different projects on eXplainable Artificial Intelligence (XAI) by the main author.

xaia

Two Instances of Interpretable Neural Network for Universal Approximations (arxiv).
This project provides two bottom-up construction of interpretable universal approximations.


Fig. left A1-3,B1-3 show perfect accuracy for function approximations of Triangular Neural Network. Fig. right shows predictions based on interpolations of neighbouring values.

gax

Augmentative eXplanation and the Distributional Gap of Confidence Optimization Score (arxiv).
This project attempts to improve predictive probability through the Augmentative Explanation process using existing heatmaps methods e.g. GradCAM and DeepLIFT.

srd

Self Reward Design with Fine-grained Interpretability (arxiv)
Simple reinforcement learning model with extreme interpretability: each neuron is purposefully designed to include semantically understandable concept.

Fig. shows trajectories taken by SRD-trained 2D robots.

xai_basic

This project provides synthetic dataset to measure the 'accuracy of heatmaps'.
Quantifying Explainability of Saliency Methods in Deep Neural Networks (arxiv).

Fig. shows the synthetic dataset.

fcalc

This project is for Generalization on the Enhancement of Layerwise Relevance Interpretability of Deep Neural Network. The project has been discontinued.