Explainable Artificial Intelligence

  1. Post-hoc Explanation Methods
    1. Model-Agnostic Feature Attribution
      1. Permutation Feature Importance
        1. Methodology and Implementation
          1. Handling Feature Dependencies
            1. Statistical Significance Testing
              1. Limitations and Pitfalls
              2. Drop-Column Feature Importance
                1. Implementation Steps
                  1. Computational Considerations
                    1. Comparison with Permutation Methods
                    2. Ablation Studies
                      1. Systematic Feature Removal
                        1. Interaction Effects
                      2. Partial Dependence Analysis
                        1. Partial Dependence Plots
                          1. Construction and Interpretation
                            1. One-Way and Two-Way Plots
                              1. Assumptions and Limitations
                              2. Individual Conditional Expectation Plots
                                1. Instance-Level Variation
                                  1. Heterogeneous Effects
                                    1. Comparison with PDP
                                    2. Accumulated Local Effects
                                      1. Handling Feature Correlation
                                        1. Unbiased Effect Estimation
                                      2. Local Explanation Methods
                                        1. LIME (Local Interpretable Model-agnostic Explanations)
                                          1. Core Principles and Algorithm
                                            1. Local Surrogate Models
                                              1. Perturbation Strategies
                                                1. Tabular Data Applications
                                                  1. Text Data Applications
                                                    1. Image Data Applications
                                                      1. Hyperparameter Selection
                                                        1. Limitations and Criticisms
                                                        2. SHAP (SHapley Additive exPlanations)
                                                          1. Game Theory Foundations
                                                            1. Shapley Value Properties
                                                              1. Efficiency, Symmetry, Dummy, Additivity
                                                                1. KernelSHAP Implementation
                                                                  1. TreeSHAP for Tree Models
                                                                    1. DeepSHAP for Neural Networks
                                                                      1. LinearSHAP for Linear Models
                                                                        1. Visualization Techniques
                                                                          1. Force Plots
                                                                            1. Summary Plots
                                                                              1. Dependence Plots
                                                                                1. Waterfall Charts
                                                                              2. Anchors
                                                                                1. High-Precision Local Rules
                                                                                  1. Coverage vs Precision Trade-offs
                                                                                    1. Rule Generation Algorithm
                                                                                      1. Beam Search Optimization
                                                                                    2. Counterfactual and Contrastive Explanations
                                                                                      1. Counterfactual Generation
                                                                                        1. Minimal Perturbation Approaches
                                                                                          1. Feasibility Constraints
                                                                                            1. Actionability Requirements
                                                                                            2. Contrastive Explanations
                                                                                              1. Foil Comparisons
                                                                                                1. Nearest Unlike Neighbor
                                                                                                2. Algorithmic Approaches
                                                                                                  1. Optimization-Based Methods
                                                                                                    1. Generative Model Approaches
                                                                                                      1. Gradient-Based Techniques
                                                                                                    2. Example-Based Explanations
                                                                                                      1. Prototypes and Criticisms
                                                                                                        1. Representative Examples
                                                                                                          1. Outlier Detection
                                                                                                          2. Influential Instances
                                                                                                            1. Training Data Attribution
                                                                                                              1. Leave-One-Out Analysis
                                                                                                              2. Case-Based Reasoning
                                                                                                                1. Similarity Metrics
                                                                                                                  1. Adaptation Strategies

                                                                                                              Previous

                                                                                                              3. Intrinsically Interpretable Models

                                                                                                              Go to top

                                                                                                              Next

                                                                                                              5. Deep Learning Specific Explanation Methods