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 Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Nonparametric Quantile Regression with ReLU-Activated Recurrent Neural Networks
Authors: Hang Yu, Lyumin Wu, Wenxin Zhou, Zhao Ren
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical experiments and an empirical study on the Dow Jones Industrial Average (DJIA) further support our theoretical findings. Numerical experiments and an empirical study on the Dow Jones Industrial Average (DJIA) further support our theoretical findings. |
| Researcher Affiliation | Academia | 1 National Key Laboratory for Novel Software Technology, Nanjing University, China 2 School of Artificial Intelligence, Nanjing University, China 3 Department of AI and Data Science, The University of Hong Kong, China 4 Department of Information and Decision Sciences, University of Illinois Chicago, USA 5 Department of Statistics, University of Pittsburgh, USA |
| Pseudocode | No | The paper does not contain any sections explicitly labeled "Pseudocode" or "Algorithm", nor does it present structured algorithm blocks. |
| Open Source Code | No | All experiments are implemented in Python. The QRF estimator is trained using the scikit-garden package, and the number of trees is set as 100. All neural networks are implemented in PyTorch. |
| Open Datasets | Yes | Numerical experiments and an empirical study on the Dow Jones Industrial Average (DJIA) further support our theoretical findings. We use DJIA data from Jan 1, 2000, to Dec 31, 2020, obtained from https://www.investing.com. We utilize U.S. GDP growth data from April 1947 to December 2024, accessible at https://fred.stlouisfed.org/series/A191RL1Q225SBEA. |
| Dataset Splits | Yes | Specifically, the dataset is split into training (80%) and validation (20%) sets. We partition the data chronologically, allocating the first 19 years for training and the final year for evaluation. The dataset is partitioned chronologically, with the most recent 30% reserved for testing. |
| Hardware Specification | No | All experiments are implemented in Python. The QRF estimator is trained using the scikit-garden package, and the number of trees is set as 100. All neural networks are implemented in PyTorch. |
| Software Dependencies | No | All experiments are implemented in Python. The QRF estimator is trained using the scikit-garden package, and the number of trees is set as 100. All neural networks are implemented in PyTorch. |
| Experiment Setup | Yes | The QRF estimator is trained using the scikit-garden package, and the number of trees is set as 100. For all NN-based estimators, we employ early stopping to mitigate overfitting... training is terminated if the validation loss does not improve for 20 consecutive epochs. For the FNN-based estimator, we adopt L = 2 hidden layers with widths W1 = W2 = 200... For the RNNand SRNN-based estimators, we use L = 3 layers with hidden width W = 100 and Re LU activation. To induce sparsity, we prune the smallest 40% of parameters and finetune the remaining weights post-training. We conduct experiments on the following two nonlinear AR models at quantile levels τ = 0.1 and τ = 0.5. For all models, the input sequence length is fixed at N = 4. |