Explainable Artificial Intelligence

  1. Deep Learning Specific Explanation Methods
    1. Gradient-Based Attribution
      1. Vanilla Gradients
        1. Gradient Computation
          1. Saliency Map Generation
            1. Limitations and Noise Issues
            2. Gradient × Input
              1. Element-wise Multiplication
                1. Attribution Magnitude
                2. Integrated Gradients
                  1. Path Integration Approach
                    1. Baseline Selection
                      1. Axiom Satisfaction
                      2. SmoothGrad
                        1. Noise Averaging Technique
                          1. Variance Reduction
                          2. Guided Backpropagation
                            1. Modified ReLU Gradients
                              1. Positive Gradient Flow
                            2. Activation-Based Methods
                              1. Class Activation Mapping
                                1. Global Average Pooling
                                  1. Linear Combination of Feature Maps
                                    1. Spatial Localization
                                    2. Grad-CAM
                                      1. Gradient-Weighted Activation Mapping
                                        1. Class-Specific Visualizations
                                          1. Multi-Layer Extensions
                                          2. Grad-CAM++
                                            1. Improved Localization
                                              1. Weighted Gradient Computation
                                              2. Score-CAM
                                                1. Perturbation-Based Scoring
                                                  1. Model-Agnostic Adaptation
                                                  2. Layer-wise Relevance Propagation
                                                    1. Relevance Conservation Principle
                                                      1. Propagation Rules
                                                        1. Deep Taylor Decomposition
                                                      2. Attention Mechanisms
                                                        1. Self-Attention Visualization
                                                          1. Attention Weight Interpretation
                                                            1. Head-Specific Analysis
                                                            2. Multi-Head Attention
                                                              1. Attention Pattern Diversity
                                                                1. Role Specialization
                                                                2. Attention Rollout
                                                                  1. Layer-wise Attention Combination
                                                                    1. Information Flow Tracking
                                                                  2. Concept-Based Explanations
                                                                    1. Testing with Concept Activation Vectors
                                                                      1. Human-Interpretable Concepts
                                                                        1. Concept Sensitivity Analysis
                                                                          1. Directional Derivatives
                                                                          2. Automated Concept Discovery
                                                                            1. Unsupervised Concept Extraction
                                                                              1. Concept Completeness
                                                                              2. Network Dissection
                                                                                1. Unit-Level Concept Alignment
                                                                                  1. Semantic Segmentation Evaluation
                                                                                2. Adversarial and Robustness Analysis
                                                                                  1. Adversarial Examples
                                                                                    1. Input Perturbation Effects
                                                                                      1. Decision Boundary Analysis
                                                                                      2. Feature Visualization
                                                                                        1. Activation Maximization
                                                                                          1. DeepDream Techniques
                                                                                            1. Feature Inversion

                                                                                        Previous

                                                                                        4. Post-hoc Explanation Methods

                                                                                        Go to top

                                                                                        Next

                                                                                        6. Tree Ensemble Specific Methods