Towards Efficient Feature Sharing in MIMO Architectures

Rémy Sun, Alexandre Ramé, Clément Masson, Nicolas Thome, Matthieu Cord

CVPRW 2022 pp. 2696-2700

doi:10.1109/CVPRW56347.2022.00303 /cvprw/2022/sun2022cvprw-efficient/

Abstract

Multi-input multi-output architectures propose to train multiple subnetworks within one base network and then average the subnetwork predictions to benefit from ensembling for free. Despite some relative success, these architectures are wasteful in their use of parameters. Indeed, we highlight in this paper that the learned subnetwork fail to share even generic features which limits their applicability on smaller mobile and AR/VR devices. We posit this behavior stems from an ill-posed part of the multi-input multi-output framework. To solve this issue, we propose a novel unmixing step in MIMO architectures that allows subnetworks to properly share features. Preliminary experiments on CIFAR 100 show our adjustments allow feature sharing and improve model performance for small architectures.

PDF CVPRW Semantic Scholar

Cite

Text

Sun et al. "Towards Efficient Feature Sharing in MIMO Architectures." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2022. doi:10.1109/CVPRW56347.2022.00303

Markdown

[Sun et al. "Towards Efficient Feature Sharing in MIMO Architectures." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2022.](https://mlanthology.org/cvprw/2022/sun2022cvprw-efficient/) doi:10.1109/CVPRW56347.2022.00303

BibTeX

@inproceedings{sun2022cvprw-efficient,
  title     = {{Towards Efficient Feature Sharing in MIMO Architectures}},
  author    = {Sun, Rémy and Ramé, Alexandre and Masson, Clément and Thome, Nicolas and Cord, Matthieu},
  booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  year      = {2022},
  pages     = {2696-2700},
  doi       = {10.1109/CVPRW56347.2022.00303},
  url       = {https://mlanthology.org/cvprw/2022/sun2022cvprw-efficient/}
}