BlockCIR
BlockCIR is a groupwise extension of ExCIR that evaluates sets of correlated features as a single entity to prevent double-counting.
Category: General AI/ML
explainable AI (XAI)
Explainable AI refers to methods and techniques that make the outputs of AI models understandable to humans.
robust centering
Robust centering involves subtracting a robust estimate, such as the median or mid-mean, from features and outputs to enhance stability in feature attribution.
Discrimination-Accuracy Optimal Classifiers
Classifiers that aim to maximize accuracy while minimizing discrimination against sensitive groups.
ExCIR
ExCIR is a correlation-aware attribution score that quantifies the impact of feature co-movement on model outputs while reducing computational costs.
Synthetic Control Arms
Control groups in clinical trials that are generated from synthetic data rather than derived from real patient populations, used to improve trial efficiency and reduce costs.
Post-Generation Selection Procedure
A method applied after data generation to refine or select the generated samples based on certain criteria, improving their statistical properties.
Generative Adversarial Network
A class of machine learning frameworks where two neural networks, a generator and a discriminator, compete against each other to produce realistic data samples.
Censoring
A statistical phenomenon where the value of an observation is only partially known, often occurring in survival analysis when subjects drop out or are lost to follow-up.
Lipschitz Function
A function is Lipschitz if there exists a constant such that the absolute difference between the function values is bounded by this constant times the distance between the inputs.