Skip to content
Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs
Python TeX
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore added figures Oct 25, 2019 note that the Brain-Score is now public Dec 3, 2019 compute r for fig. a1 Oct 28, 2019 add small yticks Dec 5, 2019
slides.pdf update slides from NeurIPS talk Dec 16, 2019

Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs

This repository hosts materials for our NeurIPS 2019 publication:

Kubilius*, Schrimpf*, et al. Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs. NeurIPS 2019 (oral)

* Equal contribution

This paper brings forward two major contributions:

  • Brain-Score, a framework for evaluating models on integrative brain measurements. Brain-Score allows to quantify how similar models are to brain responses (neural and behavioral). The current Brain-Score leaderboard is available at If you want to score your own model, use the Brain-Score repo
  • CORnet-S, a shallow recurrent artificial neural network that is the current best model on Brain-Score. A PyTorch version of ImageNet pre-trained is available at CORnet repo.

Please cite this work as follows:

  title={Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs},
  author={Kubilius, Jonas and Schrimpf, Martin and Kar, Kohitij and Hong, Ha and Majaj, Najib J and Rajalingham, Rishi and Issa, Elias B and Bashivan, Pouya and Prescott-Roy, Jonathan and Schmidt, Kailyn and Nayebi, Aran and Bear, Daniel and Yamins, Daniel L K and DiCarlo, James J},
  booktitle={Advances in Neural Information Processing Systems},

Reproducing the paper

We provide aggregated data sources for reproducing most of the figures in the paper. Run python gen_all in order to generate all figures except Fig. 4 and 5. Data for Fig. 4 involves a comparison of many models; we chose to not package all that data. For Fig. 5, run python prediction_vs_target, but note that it will be recomputed from scratch and will therefore require multiple dependencies and may take a long time.

The data used in these figures has been computed using older versions of Brain-Score and thus may not perfectly reproduce when using the latest releases. We highly recommend using the latest release of Brain-Score (and the current scores in the leaderboard at if you intend to report on your own data or models.

You can’t perform that action at this time.