Logo

T-Rex Method

A new data science framework to provide fast and robust solutions for large-scale high-dimensional settings.  T-Rex methods can be applied to diverse areas, such as, genomics, financial engineering, or signal processing. 

The Terminating-Random Experiments (T-Rex) selector is a fast and scalable framework for high-dimensional variable selection with provable finite-sample false discovery rate (FDR) control. It runs multiple early-terminated random experiments in which real variables compete with computer-generated dummy variables, and fuses the outcomes via a martingale-based calibration procedure. The T-Rex scales linearly in the number of variables and samples. This makes it applicable to problems with millions of variables. The method is easy to use and does not requirey any parameter tuning by the user.

The foundational methodology is described in:

  • J. Machkour, M. Muma, D. P. Palomar, The Terminating-Random Experiments Selector: Fast High-Dimensional Variable Selection with False Discovery Rate Control, Signal Processing, vol. 231, p. 109894, 2025. [paper]
  • J. Machkour, M. Muma, D. P. Palomar, High-Dimensional False Discovery Rate Control for Dependent Variables, Signal Processing, vol. 234, p. 109990, 2025. [paper]
  •  

Selected recent developments

  • Virtual dummies: scaling beyond the dummy-storage memory barrier. Sequential sampling of null features removes the need to store large dummy matrices, enabling T-Rex to scale to even higher-dimensional regimes while preserving its FDR-control guarantees.
    [paper] [code]
  • Learning FDR control via model-based neural networks. A learned, data-driven T-Rex calibration that improves power while maintaining approximate FDR control.
    [paper] [code]
  • Complex-Valued T-Rex (CT-Rex). Extends FDR control to complex-valued data, with applications in single-snapshot multi-source detection and direction-of-arrival (DOA) estimation.
    [paper]
  • T-Rex PCA: sparse principal component analysis with FDR control. A reformulation of sparse PCA in which loading-vector supports are obtained via the T-Rex selector. [paper]
  • Graphical models. FDR-controlled estimation of Gaussian graphical model structure using the T-Rex framework [paper]
  • Grouped variable selection. A family of T-Rex variants that  provides FDR control at the group level.  [paper]
 

A complete and up-to-date list of T-Rex publications is available on Google Scholar.