README.md

CounterFactual-XAI-Tutorials-and-Papers

CFX (CounterFactual eXplanation) is a popular method for model explanation. It is a post-hoc way that can give actionable explanations to a model result. In this repo, I will list out some SOTA related to CFX.

Notes: latest work are marked with fire 🔥

📖 Model explanation packages and libraries

DiCE [repo]

• Explaining machine learning classifiers through diverse counterfactual explanations [paper]

Alibi [repo]

• Supported methods [summary]

📚 Survey paper

• (arthur.ai) Counterfactual Explanations for Machine Learning: A Review [paper][ppt]
  • Evaluation of Counterfactual Generation Algorithms [Google Sheet]
• Benchmarking and Survey of Explanation Methods for Black Box Models [paper]

💡 Methods and works for different models

Model-Agnostic Methods

• DiCE [example code]
  • Randomized sampling
  • KD-Tree (for counterfactuals within the training data)
  • Genetic algorithm
• CFRL by Seldon Alibi [example code] [doc]
  • 🔥 Model-agnostic and Scalable Counterfactual Explanations via Reinforcement Learning [paper]

Liz's Reading list (tba)

• (Nips 2021) Designing Counterfactual Generators using Deep Model Inversion [paper]
  • No need training data
• (Nips 2021) Towards Robust and Reliable Algorithmic Recourse [paper]
  • Data shifting
  • Adversarial training
• (KDD 2021) Model-Based Counterfactual Synthesizer for Interpretation [paper]
• CounteRGAN: Generating Realistic Counterfactuals with Residual Generative Adversarial Nets [paper]
• (ICLR 2021) Counterfactual generative networks [paper][source code]
• (AAAI 2022) Amortized Generation of Sequential Algorithmic Recourses for Black-box Models[paper]
  • Algorithmic Recourses (AR) and sequential ARs
  • Reinforcement learning
• (AAAI 2022) Diverse, Global and Amortised Counterfactual Explanations for Uncertainty Estimates [paper]

🥶 on-going

• (PMLR 2020) Model-Agnostic Counterfactual Explanations for Consequential Decisions [paper] source code]
• (FAT 2020) The Hidden Assumptions Behind Counterfactual Explanations and Principal Reasons [paper]

Gradient-Based Methods

• CounterFactual (by Seldon Alibi) [example code] [doc]
  • Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPR [paper]
• CEM (By Seldon Alibi) [example code] [doc]
  • Explanations based on the Missing: Towards Contrastive Explanations with Pertinent Negatives [paper] [source code]
• DiCE
  • An explicit loss-based method [paper]
  • A Variational AutoEncoder (VAE)-based method [paper] [example code]
• 🔥 GeCo: Quality Counterfactual Explanations in Real Time [paper] [source code]

Model-Specific Methods

• Actionable Recourse in Linear Classification [paper] [source code]
• Efficient Search for Diverse Coherent Explanations [paper] [source code]

CFX on Others

• KS Test
  • 🔥 Comprehensible Counterfactual Explanation on Kolmogorov-Smirnov Test [paper] [source code]

📺 Tutorials and Discussions

• Post-hoc counterfactual generation with supervised autoencoder [ppt]
• If Only We Had Better Counterfactual Explanations: Five Key Deficits to Rectify in the Evaluation of Counterfactual XAI Techniques [paper]

👩‍🏫 Interdisciplinary Works Towards CFX

• Adquate and Fair Explanations [paper]