Photo: Cmichel67, CC BY-SA 4.0, Wikimedia Commons
Emmanuel Candès
Stanford University
- Research field
- Statistics, applied mathematics, selective inference
Emmanuel Candès and Rina Foygel Barber introduced the knockoff filter in 2015: for every predictor of a regression a fake copy is built that carries the same correlations as the original and no information about the response, the lasso runs on originals and copies together, and the copies serve as a control group that measures how often a variable with no effect would have been selected. The method keeps the false discovery rate of the selected variables at or below the chosen level in any finite sample, the first such guarantee for variable selection in regression. The unit presents the construction in four steps as the answer to the question Unit 14 left open.
In the masterclass
Section 5, after the knockoff filter and its four steps, as the answer to the selection problem of the lasso.
The research is presented in Unit 16: Multiple Testing and the False Discovery Rate, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.