Codes for the GDT algorithm on multi-task learning
Ming Yu, Varun Gupta, and Mladen Kolar. Recovery of simultaneous low rank and two-way sparse coefficient matrices, a nonconvex approach. In Electronic Journal of Statistics.
- Data_generation.m: generate true coefficient matrix
- hard_thre.m: perform (row) hard thresholding on matrix
- GDT.m: run the GDT (gradient descent with hard thresholding) algorithm
Run “Data_generation.m” to generate the data and run “GDT.m” to get the estimation.