Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Global vs Local Interpretability
Distinction between techniques explaining a model's overall behavior and those analyzing a specific prediction
Model-Agnostic Methods
Interpretation techniques that work with any type of model regardless of its architecture
Feature Importance Analysis
Identification and quantification of the relative influence of each input variable on model predictions
Interpretable Visualizations
Graphical and visual representations enabling understanding of internal workings and decisions of models
Rule-Based Methods
Extraction and generation of simple logical rules from complex models to make them understandable
Counterfactual Explanations
Generation of minimally modified examples to show how to change a prediction by exploring decision boundaries
Sensitivity Analysis
Systematic study of how variations in inputs affect model outputs to understand its robustness
Surrogate Models
Use of simple and interpretable models to approximate and explain the behavior of complex models
Feature Attribution
Distribution of responsibility for a prediction among different input features according to their contribution
Interpretability of Neural Networks
Specific techniques for understanding and visualizing the functioning of deep learning models
Fairness and Bias Detection
Systematic analysis of potential biases and discriminations in AI model decisions
Causal Interpretability
Identification of cause-and-effect relationships in models to go beyond simple correlations
Interpretability Metrics
Quantification and objective evaluation of the degree of interpretability of different models and techniques
Prototype-based Explanations
Use of representative examples to explain model decisions by comparison with typical cases
Decision Path Analysis
Tracing and visualization of the logical paths followed by the model to arrive at a specific prediction