Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Conditional Average Treatment Effect (CATE)
Measure of the variation in causal treatment effect across individuals or subgroups of the population, aiming to personalize decisions based on the expected effect for each unit.
Pearl's Collider
Structure in a causal graph where a variable is a common cause of two other variables, creating a non-causal association that must be blocked for correct inference.
Causal Fork
Structure where a single variable causes two other variables, inducing a correlation between them; this correlation is non-causal and disappears when conditioning on the common cause.
Survivorship Bias
Logical error where one focuses on subjects or entities that have passed a selection process, ignoring those that did not survive, which biases causal conclusions.
Regression Discontinuity
Quasi-experimental design that identifies causal effects by comparing observations just below and just above a treatment threshold determined exogenously.
Difference-in-Differences (DiD)
Statistical technique that estimates treatment effect by comparing the change in means over time between a treatment group and a control group.
Causal Decoupling
Principle according to which a system can be understood by decomposing it into autonomous modules, whose internal relationships are not affected by external interventions on other modules.