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..
Self-Assembling Graph Perceptrons
Authors: Jialong Chen, Tong Wang, Bowen Deng, Luonan Chen, Zibin Zheng, Chuan Chen
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Overview In the experiment, we answer three questions: first, the impact of topological structure on the perceptron model; second, the performance of SAGP on deep learning tasks; and third, the inspiration that self-assembling neural networks provide for modern deep learning. Three domains of the dataset are used: text, audio, and images. We also conducted experiments on deep graph models and temporal models. See Appendix C.1 for more dataset details. |
| Researcher Affiliation | Academia | 1Sun Yat-sen University, 2Texas A&M University, 3Shanghai Jiao Tong University, 4University of Chinese Academy of Sciences |
| Pseudocode | No | The paper describes rules like "Apoptosis Rule" and "Growth Rule" in natural language and mathematical equations, but does not present them as a structured pseudocode block or algorithm. |
| Open Source Code | Yes | 5. Open access to data and code 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: Appendix contains links to the code repository. |
| Open Datasets | Yes | Three domains of the dataset are used: text, audio, and images. We also conducted experiments on deep graph models and temporal models. See Appendix C.1 for more dataset details. ... The FSDD dataset with the smallest performance gap ... Fasion MNIST MLP-l 1024 ... CIFAR-10 MLP-l 768 ... Following the setup in [7], we compared SAGP and NDP on a toy dataset, Digit (Table 3). |
| Dataset Splits | No | The paper mentions training and validation sets being used (e.g., "epoch that achieves the lowest loss on the validation set") but does not provide specific percentages, sample counts, or a detailed splitting methodology within the provided text. It refers to Appendix C.1 for dataset details and Appendix C.2 for hyperparameters, but does not explicitly state dataset split information in the main body. |
| Hardware Specification | Yes | The only previous self-assembling model NDP was excluded from the comparison due to resource constraints (a single training run exceeding 12 hours on a single Nvidia RTX 4090). |
| Software Dependencies | No | The paper does not explicitly mention any specific software libraries or frameworks with version numbers (e.g., PyTorch, TensorFlow, Python) used for the implementation in the provided text. |
| Experiment Setup | Yes | Hyperparametes Several hyperparameters are introduced in SAGP, including L, α, γ1, γ2, β, and N. We provide a set of empirical hyperparameters, which are consistently applied across all experiments discussed in this paper (See Appendix C.2). |