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 Shift Toward Open and Reproducible AI Research," Under Review, 2026. doi: 10.48550/arXiv.2606.16974..
Grounding inductive biases in natural images: invariance stems from variations in data
Authors: Diane Bouchacourt, Mark Ibrahim, Ari Morcos
NeurIPS 2021 | Venue PDF | LLM Run Details | Input Tokens: 18,951 Total number of tokens sent to the LLM as input for this paper's analysis. | Output Tokens: 4,838 Total number of tokens produced by the LLM (including reasoning/thinking tokens) for this paper's analysis.
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We train Res Net18 on Image Net and report results on the validation set as in commonly done (e.g. in Touvron et al. [33]) since the labelled test set is not publicly available. Fixed Size Center Crop corresponds to what is usually done for augmenting validation/test images, i.e. resize the image to 256 on the shorter dimension and take a center crop of size 224. |
| Researcher Affiliation | Industry | Diane Bouchacourt , Mark Ibrahim , Ari S. Morcos Facebook AI Research EMAIL |
| Pseudocode | No | No, the paper does not contain any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | Code to reproduce the main experiments is available at https://github.com/ facebookresearch/grounding-inductive-biases. |
| Open Datasets | Yes | We train Res Net18 on Image Net and report results on the validation set as in commonly done (e.g. in Touvron et al. [33]) since the labelled test set is not publicly available. |
| Dataset Splits | Yes | We train Res Net18 on Image Net and report results on the validation set as in commonly done (e.g. in Touvron et al. [33]) since the labelled test set is not publicly available. |
| Hardware Specification | Yes | All models were trained on 8 NVIDIA V100 GPUs for 100 epochs. |
| Software Dependencies | No | No, the paper mentions the use of the PyTorch library but does not specify its version number or any other software dependencies with specific versions. |
| Experiment Setup | Yes | We use a batch size of 256 and train our models for 100 epochs... We use Adam with a learning rate of 1e-3. Default values for the distribution over s are s = 0.08, s+ = 1. |