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..
Hierarchical classification at multiple operating points
Authors: Jack Valmadre
NeurIPS 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on the i Nat21 [36] and Image Net-1k [10] datasets for image classification reveal that a naïve flat softmax classifier dominates the more elegant top-down classifiers, obtaining better accuracy at any operating point. |
| Researcher Affiliation | Academia | Jack Valmadre Australian Institute for Machine Learning, University of Adelaide EMAIL |
| Pseudocode | No | The paper describes algorithms verbally and mathematically, but does not include any formal pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is available online at https://github.com/jvlmdr/hiercls. |
| Open Datasets | Yes | Experiments on the i Nat21 [36] and Image Net-1k [10] datasets for image classification reveal that a naïve flat softmax classifier dominates the more elegant top-down classifiers, obtaining better accuracy at any operating point. |
| Dataset Splits | No | The paper mentions the 'i Nat21 validation set' but does not provide explicit percentages or sample counts for training, validation, and test splits. |
| Hardware Specification | Yes | Most experiments were conducted on a single machine with one Nvidia A6000 GPU. ... To obtain error-bars, we used a larger, shared machine with 16 Nvidia V100 GPUs (still using one GPU to train each model). |
| Software Dependencies | No | The code was implemented using the Pytorch library [28], but no specific version number for Pytorch or other software dependencies is provided. |
| Experiment Setup | Yes | For all i Nat21 experiments, we use a Res Net-18 model [19] with input images of size 224 224. We start from the Pytorch Image Net-pretrained checkpoint [28] and train for 20 epochs using SGD with momentum 0.9, cosine schedule [24], batch size 64, initial learning rate 0.01 and weight decay 0.0003. |