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..
SymMaP: Improving Computational Efficiency in Linear Solvers through Symbolic Preconditioning
Authors: Hong Wang, Jie Wang, Minghao Ma, Haoran Shao, Haoyang Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conducted comprehensive experiments to evaluate the Sym Ma P framework, organized into three primary sections: 1. Assessment of three different preconditioners and optimization goals across various datasets to determine the effectiveness of Sym Ma P, 2. Analysis of associated computational cost and the interpretability of the learned symbolic expressions, 3. Ablation studies. |
| Researcher Affiliation | Academia | Hong Wang1,2,3 , Jie Wang1,2,3 , Minghao Ma1, Haoran Shao1, Haoyang Liu1,2,3 1 University of Science and Technology of China 2 CAS Key Laboratory of Technology in GIPAS, University of Science and Technology of China 3 Mo E Key Laboratory of Brain-inspired Intelligent Perception and Cognition, University of Science and Technology of China EMAIL, EMAIL |
| Pseudocode | Yes | The detailed steps are as follows and pseudocode is provided in the Appendix C. ... Algorithm 1: RNN-based Symbolic Discovery Process ... Algorithm 2: Deep Symbolic Optimization for Matrix Preconditioning Parameter |
| Open Source Code | Yes | 1Our code is available at https://github.com/Minghaom2/Sym Ma P. |
| Open Datasets | No | The generated dataset and training time are available in Appendix D.6. |
| Dataset Splits | Yes | This process yields the required training dataset, where each data point contains: 1. the problem feature parameters xi. 2. the optimal preconditioning parameters yi. i = 1, 2, . . . , n, and n is the number of data, typically set to n = 1200, with 1000 allocated for the training set and 200 for the test set. |
| Hardware Specification | Yes | Environment (Env1): Platform: Windows11 version 22631.4169, WSL Operating System: Ubuntu 22.04.3 CPU Processor: AMD Ryzen 9 5900HX with Radeon Graphics CPU, clocked at 3.30GHz 2. Environment (Env2): Platform & Operating System: Ubuntu 18.04.4 LTS CPU Processor: Intel(R) Xeon(R) Gold 6246R CPU at 3.40GHz GPU Processor: Ge Force RTX 3090 24GB |
| Software Dependencies | No | Library: CUDA Version 11.3 |
| Experiment Setup | Yes | Table 7: Hyperparameters of Sym MAP (Default Model) Hyperparameter Value Number of LSTM layers 1 Number of LSTM units 32 Number of training samples 2,000,000 Batch size 1,000 Risk factor ε 0.05 Minimal expression length 4 Maximal expression length 64 Learning rate 0.0005 Weight of entropy regularization 0.03 |