False Discovery Rate

False Discovery Rate (FDR) control addresses the challenge of minimizing false positives in large-scale multiple hypothesis testing, aiming to maintain a specified proportion of false discoveries among all rejected null hypotheses. Current research focuses on developing adaptive and efficient FDR control methods across diverse settings, including online testing, high-dimensional data (e.g., using sparse PCA and Gaussian graphical models), and incorporating auxiliary information or complex model structures (e.g., via deep learning or conditional prediction functions). These advancements are crucial for reliable scientific inference across various fields, from A/B testing and neuroimaging to genomics and anomaly detection in time series, enabling more powerful and trustworthy conclusions from large datasets.