TrIM: Transformed Iterative Mondrian Forests for Gradient-based Dimension Reduction and High-Dimensional Regression
Researchers have developed a new algorithm called TrIM (Transformed Iterative Mondrian Forests) designed for efficient gradient-based linear dimension reduction and high-dimensional regression. This method utilizes a Mondrian forest to identify relevant feature subspaces by estimating the expected gradient outer product (EGOP). The TrIM algorithm iteratively refines the Mondrian forest by updating features and weights based on the EGOP estimate, offering theoretical guarantees for consistency and convergence rates. AI