Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Online Bootstrap Confidence Intervals for the Stochastic Gradient Descent Estimator
Authors: Yixin Fang, Jinfeng Xu, Lei Yang
JMLR 2018 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The finite-sample performance and numerical utility is evaluated by simulation studies and real data applications. Keywords: Bootstrap, Interval estimation, Generalized linear models, Large datasets, M-estimators, Quantile regression, Resampling methods, Stochastic gradient descent. In Section 4, we demonstrate the performance of the proposed procedures in finite samples via simulation studies and three real data applications. |
| Researcher Affiliation | Academia | Yixin Fang EMAIL Department of Mathematical Sciences New Jersey Institute of Technology. Jinfeng Xu EMAIL Department of Statistics and Actuarial Science The University of Hong Kong. Lei Yang EMAIL Department of Population Health New York University School of Medicine. |
| Pseudocode | No | The paper describes the stochastic gradient descent (SGD) and perturbed SGD updates using mathematical equations (e.g., equations 3, 5, 17, 18, 20, 21, 24, 25, 29, 30) which define recursive procedures. However, there are no explicitly labeled pseudocode blocks or algorithms in the paper. |
| Open Source Code | No | The paper provides a license link for the paper itself: "License: CC-BY 4.0, see https://creativecommons.org/licenses/by/4.0/. Attribution requirements are provided at http://jmlr.org/papers/v19/17-370.html." There is no explicit statement or link in the paper indicating that the authors have released or made available the source code for the methodology described. |
| Open Datasets | Yes | All the three datasets are publicly available on UCI machine learning repository. The POWER dataset contains 2,075,259 observations... The SKIN dataset contains 245,057 observations... The GAS dataset contains 919,438 observations... |
| Dataset Splits | No | The paper mentions excluding initial estimates for calculations: "When we calculate the average SGD estimators (4) and (6), the first 2000 and 4000 estimates are excluded for N = 10,000 and N = 20,000, respectively." This refers to excluding early iterations from averaging, not specifying standard training/test/validation dataset splits. No other specific dataset split information is provided for reproduction. |
| Hardware Specification | No | The paper discusses computational and memory efficiency of SGD methods but does not provide specific details about the hardware (e.g., GPU/CPU models, memory specifications) used to conduct the simulation studies or real data applications. |
| Software Dependencies | No | The paper mentions "standard softwares such as SAS and R" as tools that can fit linear and logistic regression to datasets, but it does not specify the versions of these or any other software packages used in their own experimental implementation. There are no specific library names with version numbers provided. |
| Experiment Setup | Yes | We consider the learning rate α = 2/3. For each data repetition, we use Wnb exp(1) as random weights and generate B = 200 copies of random weights whenever a new data point is read. Then, for each data repetition, we obtain the SGD estimate (4), and apply the following procedures to construct 95% confidence intervals... When we calculate the average SGD estimators (4) and (6), the first 2000 and 4000 estimates are excluded for N = 10,000 and N = 20,000, respectively. |