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..
A Minimalist Example of Edge-of-Stability and Progressive Sharpening
Authors: Liming Liu, Zixuan Zhang, Simon S Du, Tuo Zhao
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments are simple numerical simulations designed to verify our theoretical results and illustrate the properties of our setting. (From NeurIPS Paper Checklist - Limitations section, Justification) Also, Figure 1: Set λ1 = 100 and λ2 = 0.01. We train our model with η = 1/20 for 10000 iterations. |
| Researcher Affiliation | Academia | Liming Liu Georgia Institute of Technology EMAIL Zixuan Zhang Georgia Institute of Technology EMAIL Simon Shaolei Du University of Washington EMAIL Tuo Zhao Georgia Institute of Technology EMAIL |
| Pseudocode | No | The paper describes the methodology using prose and mathematical equations but does not contain any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | Our experiments are all numerical simulations, so no specific dataset is required. If the paper is accepted, we will provide our code, although it is very simple. (From NeurIPS Paper Checklist - Open access to data and code section, Justification) |
| Open Datasets | Yes | We visualize part of the GD trajectory with different learning rates, as well as the Gradient Flow (GF) trajectory, while training a four-layer MLP on a binary classification problem using the CIFAR10 dataset. |
| Dataset Splits | No | The paper mentions training on the CIFAR10 dataset but does not explicitly provide specific training/test/validation splits or reference predefined splits for this dataset. |
| Hardware Specification | Yes | Our numerical simulations can be run directly on a CPU and do not require any special hardware. (From NeurIPS Paper Checklist - Experiments compute resources section, Justification) |
| Software Dependencies | No | The paper does not provide specific software dependencies or version numbers for any libraries or tools used in the numerical simulations. |
| Experiment Setup | Yes | Figure 1: Set λ1 = 100 and λ2 = 0.01. We train our model with η = 1/20 for 10000 iterations. We initialize the weights within the initialization set X(η). |