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..
Interpretable and Parameter Efficient Graph Neural Additive Models with Random Fourier Features
Authors: Thummaluru Siddartha Reddy, Vempalli Naga Sai Saketh, Mahesh Chandran
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we evaluate G-NAMRFF performance and demonstrate its ability to learn interpretable representations on node and graph classification tasks. ... In Table 1, we report mean classification accuracy along with the standard deviation over five independent runs with different random seeds. ... In Table 2, we present the mean graph classifica- tion accuracies from 10-fold cross-validation, each repeated over three random seeds. ... In this section, we analyze the impact of hyperparameters on the model performance. |
| Researcher Affiliation | Industry | Fujitsu Research of India, Bangalore. EMAIL |
| Pseudocode | Yes | Algorithm 1 Algorithm for G-NAMRFF |
| Open Source Code | Yes | Code to reproduce the results is available at https://github.com/Fujitsu Research/GNAM-RFF |
| Open Datasets | Yes | To begin with, we analyze the univariate functions learnt on Pub Med [29] dataset with nodeclassification task. ... we empirically evaluate its performance on medium-scale and large scale datasets namely Cora, Citeseer, Pub Med [36], Cornell[23], ogbn-arxiv and ogbn-products [16]. ... we discuss the interpretability on the graph classification task with the widely studied Mutagenicity dataset [18, 9]. |
| Dataset Splits | Yes | For node classification tasks, we follow the standard dataset splits as specified in [16, 20, 23]. In the case of graph classification tasks, where no standard splits are available, we employ 10-fold cross-validation. |
| Hardware Specification | Yes | We have conducted all the experiments using an NVIDIA A30 GPU. ... All the details regarding the compute workers and runtime plots are mentioned in the Appendix. |
| Software Dependencies | No | All models are trained for 1000 epochs using the Adam W optimizer with randomly initialized parameters. |
| Experiment Setup | Yes | Hyperparameter details (e.g., number of RFFs (M), scale (Θ), filter order (R), and optimizer settings) are provided in the Appendix C.1. ... All models are trained for 1000 epochs using the Adam W optimizer with randomly initialized parameters. ... The specific learning rates and weight decay values used across different experiments are also reported in Table 3. |