ICML 2020poster178 citations

From ImageNet to Image Classification: Contextualizing Progress on Benchmarks

Dimitris Tsipras, Shibani Santurkar, Logan Engstrom, Andrew Ilyas, Aleksander Madry

Abstract

Building rich machine learning datasets in a scalable manner often necessitates a crowd-sourced data collection pipeline. In this work, we use human studies to investigate the consequences of employing such a pipeline, focusing on the popular ImageNet dataset. We study how specific design choices in the ImageNet creation process impact the fidelity of the resulting dataset—including the introduction of biases that state-of-the-art models exploit. Our analysis pinpoints how a noisy data collection pipeline can lead to a systematic misalignment between the resulting benchmark and the real-world task it serves as a proxy for. Finally, our findings emphasize the need to augment our current model training and evaluation toolkit to take such misalignment into account.

BibTeX
@InProceedings{pmlr-v119-tsipras20a,
  title = 	 {From {I}mage{N}et to Image Classification: Contextualizing Progress on Benchmarks},
  author =       {Tsipras, Dimitris and Santurkar, Shibani and Engstrom, Logan and Ilyas, Andrew and Madry, Aleksander},
  booktitle = 	 {Proceedings of the 37th International Conference on Machine Learning},
  pages = 	 {9625--9635},
  year = 	 {2020},
  editor = 	 {III, Hal Daumé and Singh, Aarti},
  volume = 	 {119},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {13--18 Jul},
  publisher =    {PMLR},
  pdf = 	 {http://proceedings.mlr.press/v119/tsipras20a/tsipras20a.pdf},
  url = 	 {https://proceedings.mlr.press/v119/tsipras20a.html},
  abstract = 	 {Building rich machine learning datasets in a scalable manner often necessitates a crowd-sourced data collection pipeline. In this work, we use human studies to investigate the consequences of employing such a pipeline, focusing on the popular ImageNet dataset. We study how specific design choices in the ImageNet creation process impact the fidelity of the resulting dataset—including the introduction of biases that state-of-the-art models exploit. Our analysis pinpoints how a noisy data collection pipeline can lead to a systematic misalignment between the resulting benchmark and the real-world task it serves as a proxy for. Finally, our findings emphasize the need to augment our current model training and evaluation toolkit to take such misalignment into account.}
}