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..

Infinite Neural Operators: Gaussian processes on functions

Authors: Daniel Augusto de Souza, Yuchen Zhu, Jake Cunningham, Yuri F. Saporito, Diego Mesquita, Marc Deisenroth

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments reinforce our theoretical results, showcasing the agreement between increasingly wide NOs and our derived expressions for the infinite limit at initialization. Additionally, we compare the performance of these models in a regression setting.
Researcher Affiliation Academia Daniel Augusto de Souza University College London Yuchen Zhu University College London Harry Jake Cunningham University College London Yuri Saporito Fundac ao Getulio Vargas Diego Mesquita Fundac ao Getulio Vargas Marc Peter Deisenroth University College London
Pseudocode No The paper describes theoretical results, mathematical derivations, and experimental validations, but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes Code is avaliable at https://github.com/spectraldani/infinite-neural-operator.
Open Datasets Yes 1D Burgers equation dataset from Takamoto et al. (2022)... The original data can be downloaded at https://darus.uni-stuttgart.de/api/access/datafile/268193.
Dataset Splits Yes We evaluate all models using 5-fold crossvalidation and report the average and standard deviation of the empirical L2 norm of the prediction error.
Hardware Specification Yes All experiments were run in a desktop machine with a 3.8GHz Intel Core i7-9800X CPU and a 24GB NVIDIA Titan RTX (TU102) GPU.
Software Dependencies No All experiments were implemented in Python, mainly based on the GPy Torch (Gardner et al., 2018) library, and run in a desktop computer using a Titan RTX. No specific version numbers for Python or GPy Torch are provided.
Experiment Setup Yes The hyperparameters of the -FNO are estimated using L-BFGS, while the parameters of the FNOs are optimized with Adam using a step size of 0.001.