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..
RMSprop converges with proper hyper-parameter
Authors: Naichen Shi, Dawei Li, Mingyi Hong, Ruoyu Sun
ICLR 2021 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We provide empirical evidence for such a phase transition in our numerical experiments. |
| Researcher Affiliation | Academia | IOE, University of Michigan, EMAIL. Part of the work was done when Naichen Shi was working with Prof. Ruoyu Sun as an intern. ISE, University of Illinois at Urbana-Champaign. EMAIL ECE, University of Minnesota Twin Cities, EMAIL. University of Illinois at Urbana-Champaign. EMAIL. |
| Pseudocode | Yes | Algorithm 1 Randomly Shuffled Adam |
| Open Source Code | Yes | All codes generating experimental results are available on the Github repository https://github. com/soundsinteresting/RMSprop |
| Open Datasets | Yes | We conduct experiments of image classification and GAN on MNIST and CIFAR-10 to support our theoretical findings. |
| Dataset Splits | No | The paper mentions training and test sets but does not explicitly describe validation dataset splits. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used, such as CPU or GPU models. |
| Software Dependencies | No | The paper does not list specific software dependencies with version numbers. |
| Experiment Setup | Yes | In the CIFAR experiments, we use Res Net-18. We choose β2 = 0.8, 0.9, 0.95, 0.99 respectively. With different batch sizes 8, 16, 32, we run each algorithm for 100 epochs without explicit regularization. |