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..
Sharper Convergence Rates for Nonconvex Optimisation via Reduction Mappings
Authors: Evan Markou, Thalaiyasingam Ajanthan, Stephen Gould
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | G Experimental Validation of Results. To validate our theoretical findings, we present three illustrative examples. These examples demonstrate the application and performance of our proposed reduction mapping method in different optimisation scenarios, from a simple 2D quadratic case to a more complex high-dimensional nonlinear problem. |
| Researcher Affiliation | Collaboration | Evan Markou Australian National University EMAIL Thalaiyasingam Ajanthan Australian National University & Pluralis Research EMAIL Stephen Gould Australian National University EMAIL |
| Pseudocode | Yes | Below, we make explicit the form of the geometrically preconditioned gradient descent algorithm applied to the reduced problem F under the pullback metric induced by Φ x(t+1) 1 = x(t) 1 η R 1 F x(t) 1 , with R := D Φ D Φ . (Geo Prec GD) |
| Open Source Code | Yes | Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: We provide sufficient instructions to reproduce the results. |
| Open Datasets | No | The paper discusses synthetic experiments in Appendix G, where objective functions are defined (e.g., G(x, y) = x^2 + 10(y - x)^2). It does not reference or provide access information for any pre-existing public datasets. |
| Dataset Splits | No | The paper uses synthetic data for its experimental validation, with functions explicitly defined within Appendix G for different optimization scenarios. As such, there are no traditional training, validation, or test dataset splits in the context of pre-existing datasets. |
| Hardware Specification | No | The paper does not explicitly describe the hardware used for running its experiments. Appendix G, which details the experimental validation, focuses on the mathematical formulation of the problems and algorithms, without mentioning specific GPU or CPU models, memory, or other hardware specifications. |
| Software Dependencies | No | The paper does not provide specific ancillary software details with version numbers. While experiments are mentioned, there are no explicit listings of programming languages, libraries, or solvers and their versions used for implementation. |
| Experiment Setup | Yes | For example, in Section G.1, it states: 'The learning rates for each method are set based on the smoothness constant of the function, i.e., η = 1/β.' In Section G.2 and G.3, it mentions: 'The learning rates for each method are set based on Armijo’s backtracking line search.' It also specifies problem dimensions, such as n=40 for high-dimensional problems. |