ML Anthology

Optimizing Rank-Based Metrics with Blackbox Differentiation

Michal Rolinek, Vit Musil, Anselm Paulus, Marin Vlastelica, Claudio Michaelis, Georg Martius
CVPR 2020
doi:10.1109/CVPR42600.2020.00764 /cvpr/2020/rolinek2020cvpr-optimizing/

Abstract

Rank-based metrics are some of the most widely used criteria for performance evaluation of computer vision models. Despite years of effort, direct optimization for these metrics remains a challenge due to their non-differentiable and non-decomposable nature. We present an efficient, theoretically sound, and general method for differentiating rank-based metrics with mini-batch gradient descent. In addition, we address optimization instability and sparsity of the supervision signal that both arise from using rank-based metrics as optimization targets. Resulting losses based on recall and Average Precision are applied to image retrieval and object detection tasks. We obtain performance that is competitive with state-of-the-art on standard image retrieval datasets and consistently improve performance of near state-of-the-art object detectors.

PDF CVPR Semantic Scholar

Cite

Text

Rolinek et al. "Optimizing Rank-Based Metrics with Blackbox Differentiation." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020. doi:10.1109/CVPR42600.2020.00764

Markdown

[Rolinek et al. "Optimizing Rank-Based Metrics with Blackbox Differentiation." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.](https://mlanthology.org/cvpr/2020/rolinek2020cvpr-optimizing/) doi:10.1109/CVPR42600.2020.00764

BibTeX

@inproceedings{rolinek2020cvpr-optimizing,
  title     = {{Optimizing Rank-Based Metrics with Blackbox Differentiation}},
  author    = {Rolinek, Michal and Musil, Vit and Paulus, Anselm and Vlastelica, Marin and Michaelis, Claudio and Martius, Georg},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year      = {2020},
  doi       = {10.1109/CVPR42600.2020.00764},
  url       = {https://mlanthology.org/cvpr/2020/rolinek2020cvpr-optimizing/}
}