Transfer Learning for High-dimensional Quantile Regression with Statistical Guarantee

Authors: Sheng Qiao, Yong He, Wenxin Zhou

TMLR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Thorough simulation studies justify our theoretical analysis.
Researcher Affiliation Academia Sheng Qiao EMAIL Department of Mathematics University of California San Diego San Diego, CA 92093, USA Yong He EMAIL Institute for Financial Studies Shandong University Jinan, 250100, China Wen-Xin Zhou EMAIL Department of Information and Decision Sciences University of Illinois at Chicago Chicago, IL 60607, USA
Pseudocode Yes Algorithm 1: Oracle ℓ1-Trans-SQR Algorithm 2: Transferable Source Detection Algorithm 3: Trans-SQR Algorithm 4: Oracle Trans-SQR with ℓ0-norm constrained transferring set
Open Source Code No The paper does not contain any explicit statements about providing open-source code, nor does it include links to code repositories.
Open Datasets No The paper conducts numerical studies using simulated data, explicitly describing the data generation process (e.g., "The covariates from target x(0) i are i.i.d. Gaussian with mean zero and covariance matrix Σ...", "ϵ(0) i are i.i.d. Gaussian with mean zero and variance one"). It does not use or provide access information for any pre-existing public datasets.
Dataset Splits No The numerical studies are based on simulated data, and while Algorithm 2 describes partitioning target data into 'q' subsets for transferable source detection (a form of cross-validation), the paper does not specify fixed training/test/validation splits for a specific dataset (either real or generated) in the traditional sense required for reproducing the main model evaluation.
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments or simulations.
Software Dependencies No The paper mentions statistical methods and estimators (e.g., "convolution-type smoothed quantile regression", "iteratively reweighted ℓ1-penalized SQR estimator") but does not list any specific software libraries, frameworks, or programming languages with version numbers used for implementation.
Experiment Setup Yes We consider p = 500, n0 = 200, and n1, . . . , n10 = 150. The covariates from target x(0) i are i.i.d. Gaussian with mean zero and covariance matrix Σ with Σjj = 0.5|j j |... ϵ(0) i are i.i.d. Gaussian with mean zero and variance one... For the target, the true parameter β , we set s = 5, βj = 0.5 for j {1, . . . , s}, and βj = 0 otherwise. ... We take λω = Cω p log(p)/(n Am + n0), λδ = Cδ p log(p)/n0, where Cω and Cδ are sufficiently large constants...