Global Explanation

Global explanation in machine learning aims to provide a comprehensive understanding of a model's overall behavior, rather than just explaining individual predictions. Current research focuses on developing methods that generate global explanations across various model architectures, including graph neural networks and deep learning models for image and point cloud data, often employing techniques like counterfactual reasoning, Shapley values, and activation maximization. These advancements are crucial for increasing trust and transparency in AI systems, particularly in high-stakes applications like healthcare and autonomous driving, by enabling better model debugging, bias detection, and ultimately, more reliable decision-making.

Papers

October 21, 2022

October 13, 2022

Global Explainability of GNNs via Logic Combination of Learned Concepts
Steve Azzolin, Antonio Longa, Pietro Barbiero, Pietro Liò, Andrea Passerini
Ground Truth Gene Level GNN Concept Learning GNN Explanation Global Explanation

October 10, 2022

September 22, 2022

Concept Activation Regions: A Generalized Framework For Concept-Based Explanations
Jonathan Crabbé, Mihaela van der Schaar
Latent Space Concept Activation Vector Concept Based Explanation Global Explanation Generalized Framework

August 22, 2022

Global Concept-Based Interpretability for Graph Neural Networks via Neuron Analysis
Han Xuanyuan, Pietro Barbiero, Dobrik Georgiev, Lucie Charlotte Magister, Pietro Lió
Graph Neural Network Inherent Interpretability World Event Concept Based GNN Explanation Global Explanation Neuron Level

August 12, 2022

Unifying local and global model explanations by functional decomposition of low dimensional structures
Munir Hiabu, Joseph T. Meyer, Marvin N. Wright
Global Explanation Low Dimensional Structure Partial Dependence Plot Function Decomposition

June 17, 2022

Explainable Global Error Weighted on Feature Importance: The xGEWFI metric to evaluate the error of data imputation and data augmentation
Jean-Sébastien Dessureault, Daniel Massicotte
Data Augmentation Feature Importance Data Imputation Novel Metric Global Explanation

March 30, 2022

Towards Interpretable Deep Reinforcement Learning Models via Inverse Reinforcement Learning
Sean Xie, Soroush Vosoughi, Saeed Hassanpour
Reinforcement Learning Deep Learning Inherent Interpretability Inverse Reinforcement Learning Global Explanation Reinforcement Learning Model Adversarial Inverse Reinforcement Learning

March 17, 2022

Visualizing Global Explanations of Point Cloud DNNs
Hanxiao Tan
Point Cloud Explainability Method Global Explanation Explanation Based

January 12, 2022

SLISEMAP: Supervised dimensionality reduction through local explanations
Anton Björklund, Jarmo Mäkelä, Kai Puolamäki
Black Box Model Local Explanation Local Interpretable Model Agnostic Explanation Global Explanation Supervised Dimensionality Reduction

December 17, 2021

Global explainability in aligned image modalities
Justin Engelmann, Amos Storkey, Miguel O. Bernabeu
Global Explanation Image Modality Ultra Wide Pixel Wise Aggregation

Global Explanation

Papers

Integrating Policy Summaries with Reward Decomposition for Explaining Reinforcement Learning Agents

Global Counterfactual Explainer for Graph Neural Networks

Global Explainability of GNNs via Logic Combination of Learned Concepts

DALE: Differential Accumulated Local Effects for efficient and accurate global explanations

Local Interpretable Model Agnostic Shap Explanations for machine learning models

Concept Activation Regions: A Generalized Framework For Concept-Based Explanations

Global Concept-Based Interpretability for Graph Neural Networks via Neuron Analysis

Unifying local and global model explanations by functional decomposition of low dimensional structures

Explainable Global Error Weighted on Feature Importance: The xGEWFI metric to evaluate the error of data imputation and data augmentation

Towards Interpretable Deep Reinforcement Learning Models via Inverse Reinforcement Learning

Visualizing Global Explanations of Point Cloud DNNs

SLISEMAP: Supervised dimensionality reduction through local explanations

Global explainability in aligned image modalities