Sparse Nonlinear Regression: Parameter Estimation under Nonconvexity

Authors: Zhuoran Yang, Zhaoran Wang, Han Liu, Yonina Eldar, Tong Zhang

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Detailed numerical results are provided to back up our theory.
Researcher Affiliation Academia Zhuoran Yang ZY6@PRINCETON.EDU Zhaoran Wang ZHAORAN@PRINCETON.EDU Han Liu HANLIU@PRINCETON.EDU Department of Operations Research and Financial Engineering, Princeton University, Princeton, NJ 08544, USA Yonina C. Eldar YONINA@EE.TECHNION.AC.IL Department of EE Technion, Israel Institute of Technology, Haifa 32000, Israel Tong Zhang TZHANG@STAT.RUTGERS.EDU Department of Statistics, Rutgers University, Piscataway, New Jersey 08854, USA
Pseudocode Yes Algorithm 1 Proximal gradient algorithm for solving the 1-regularized problem in (1.2). Algorithm 2 The Barzilai-Borwein (BB) spectral approach for choosing αt in Line 1 of Algorithm 1.
Open Source Code No The paper describes algorithms and numerical experiments but does not provide any explicit statement or link for the availability of its source code.
Open Datasets Yes To show the effectiveness of the proposed method, we study the Computer Audition Lab 500-Song (CAL500) dataset (Turnbull et al., 2008), which can be obtained from the publicly available Mulan data library (Tsoumakas et al., 2011).
Dataset Splits Yes For the linear framework we apply the 1-regularized regresion (Lasso) (Tibshirani, 1996). ... The optimization problem of Lasso is also solved using Algorithm 1. We plot the 2-errors of these two techniques against the effective sample size in Figure 1-(c), which shows that the proposed method outperforms the linear approach. ... the regularization parameter of Lasso is selected via 5-fold cross-validation.
Hardware Specification No The paper discusses running numerical experiments but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for these computations.
Software Dependencies No The paper mentions the use of 'Lasso' and references algorithms like 'ISTA' and 'Spa RSA method' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes The parameters of Algorithm 1 are chosen as αmin = 1/αmax = 1030, η = 2, M = 5, and ζ = tol = 10 5. The 2-errors reported are based on 100 independent experiments. We plot the 2-errors against the effective sample size log d/n in Figure 1. The figure illustrates that β β 2 grows sublinearly with