Revisiting Stein's Paradox: Multi-Task Averaging

Authors: Sergey Feldman, Maya R. Gupta, Bela A. Frigyik

JMLR 2014 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Simulations and real data experiments demonstrate the advantage of the proposed MTA estimators over standard averaging and James-Stein estimation. Keywords: multi-task learning, James-Stein, Stein s paradox
Researcher Affiliation Collaboration Sergey Feldman EMAIL Data Cowboys 9126 23rd Ave. NE Seattle, WA 98115, USA Maya R. Gupta EMAIL Google 1225 Charleston Rd Mountain View, CA 94301, USA Bela A. Frigyik EMAIL Institute of Mathematics and Informatics University of P ecs H-7624 P ecs, Ifj us ag St. 6, Hungary
Pseudocode No The paper describes the methodology using mathematical formulations and detailed prose, but it does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Research-grade Matlab code and the data used in these experiments can be found at http://mayagupta.org/publications.html.
Open Datasets Yes Research-grade Matlab code and the data used in these experiments can be found at http://mayagupta.org/publications.html.
Dataset Splits Yes For the cross-validated versions, we randomly subsampled Nt/2 samples and chose the value of γ for MTA Constant/Minimax or λ for James-Stein that resulted in the lowest average left-out risk compared to the sample mean estimated with all Nt samples. ... For the cross-validation estimators, we cross-validate the regularization parameter from the set {2 15, 2 14, . . . , 214, 215}. ... Cross-validation parameters were chosen using double-leave-one-out cross-validation (for each sample left out for test, the remaining N-1 samples undergo leave-one-out cross-validation to optimize (23)).
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory, or specific computing environments) used for running the experiments.
Software Dependencies No Research-grade Matlab code and the data used in these experiments can be found at http://mayagupta.org/publications.html. (Mention of Matlab without a version number is insufficient.)
Experiment Setup Yes We used the following parameters for CV: γ {2 5, 2 4, . . . , 25} for the MTA estimators and for cross-validated James-Stein a comparable set of λ spanning (0, 1) by the transformation λ = γ γ+1. ... For the cross-validation estimators, we cross-validate the regularization parameter from the set {2 15, 2 14, . . . , 214, 215}. ... For each experiment, a single pooled variance estimate when needed was used for all tasks: σ2 t = σ2, for all t.