Knockoffs-based False Discovery Rate Control and Simplification for Deep Neural Networks
Researchers have developed new methods for simplifying deep neural networks by controlling false discovery rates. These techniques aim to reduce computational complexity and cost by identifying and removing irrelevant input variables. The proposed methods, including a one-layer filter, a multiple-layer filter, and a variable weight aggregation filter, build upon existing knockoff methods and regularized neural networks. AI
IMPACT These variable screening methods could lead to more efficient and less computationally expensive deep learning models.