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..
Causal-R: A Causal-Reasoning Geometry Problem Solver for Optimized Solution Exploration
Authors: Wenjun Wu, Lingling Zhang, Bo Zhao, Muye Huang, QianYing Wang, Jun Liu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments demonstrate the effectiveness of our method to obtain multiple shorter and interpretable solutions. |
| Researcher Affiliation | Collaboration | 1School of Computer Science and Technology, Xi an Jiaotong University, Xi an, 710049, China 2Lenovo Research |
| Pseudocode | Yes | The corresponding algorithm is given in appendix A. |
| Open Source Code | Yes | Reference of codes is available at https://github.com/nicktech-git/Causal-R. |
| Open Datasets | Yes | Following previous symbolic-based methods, we choose the popularly used Geometry3K 19 as our benchmark, which includes 3,002 geometry problems that covers a wide range of problem types such as triangles, circles and polygons. |
| Dataset Splits | Yes | For use of neural models, it is split into 2,101, 300 and 601 samples for training, validation and testing respectively. |
| Hardware Specification | Yes | The machine is 24GB NVIDIA Ge Force RTX 3090. |
| Software Dependencies | No | The paper mentions using 'Python dictionaries' for a basic strategy in Appendix C, but does not provide specific version numbers for any software, libraries, or dependencies used in the experiments. |
| Experiment Setup | Yes | The maximum constraints of deduction iterations and number of candidate solution paths w.r.t. each target are set to 7 and {1,2,3}, respectively. The early stopping mechanism is used to terminate the iterative deduction process once all target nodes have been reached and the number of iterations exceeds 4. |