Reproducibility Index

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..

Spatial Mixture-of-Experts

Authors: Nikoli Dryden, Torsten Hoefler

NeurIPS 2022 | Venue PDF | LLM Run Details

Reproducibility Variable	Result	LLM Response
Research Type	Experimental	We conduct experiments on several benchmark datasets with SMOEs and conduct extensive ablation studies of SMOE design decisions ( 3).
Researcher Affiliation	Academia	Nikoli Dryden ETH Zürich EMAIL Torsten Hoeﬂer ETH Zürich EMAIL
Pseudocode	No	The paper describes the SMOE layer and training processes conceptually and mathematically but does not include any pseudocode or algorithm blocks.
Open Source Code	Yes	Our code is available at https://github.com/spcl/smoe.
Open Datasets	Yes	Weather Bench [78], ENS-10 dataset [8], Image Net-1k [24]. The paper also states: "Most datasets we use are publicly available."
Dataset Splits	Yes	We use the data subset suggested by Rasp et al. [78] at 5.625 resolution (32 64 grid points) and train on data from 1979 2015, validate on 2016, and report test results for 2017 2018.
Hardware Specification	Yes	All results were run using Py Torch [73] version 1.11 on a large cluster with 16 GB V100 GPUs.
Software Dependencies	Yes	All results were run using Py Torch [73] version 1.11 on a large cluster with 16 GB V100 GPUs.
Experiment Setup	Yes	All models were trained with batch size 32, Adam [54] with a learning rate of 0.001 (decayed by 10 after no validation improvement for 15 epochs), and early stopping after no improvement for 30 epochs.