English

A Closer Look at Branch Classifiers of Multi-exit Architectures

Computer Vision and Pattern Recognition 2022-07-14 v2

Abstract

Multi-exit architectures consist of a backbone and branch classifiers that offer shortened inference pathways to reduce the run-time of deep neural networks. In this paper, we analyze different branching patterns that vary in their allocation of computational complexity for the branch classifiers. Constant-complexity branching keeps all branches the same, while complexity-increasing and complexity-decreasing branching place more complex branches later or earlier in the backbone respectively. Through extensive experimentation on multiple backbones and datasets, we find that complexity-decreasing branches are more effective than constant-complexity or complexity-increasing branches, which achieve the best accuracy-cost trade-off. We investigate a cause by using knowledge consistency to probe the effect of adding branches onto a backbone. Our findings show that complexity-decreasing branching yields the least disruption to the feature abstraction hierarchy of the backbone, which explains the effectiveness of the branching patterns.

Keywords

Cite

@article{arxiv.2204.13347,
  title  = {A Closer Look at Branch Classifiers of Multi-exit Architectures},
  author = {Shaohui Lin and Bo Ji and Rongrong Ji and Angela Yao},
  journal= {arXiv preprint arXiv:2204.13347},
  year   = {2022}
}

Comments

under consideration at Computer Vision and Image Understanding

R2 v1 2026-06-24T11:01:11.552Z