Summary of Early-exit Convolutional Neural Networks, by Edanur Demir et al.
Early-exit Convolutional Neural Networks
by Edanur Demir, Emre Akbas
First submitted to arxiv on: 9 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach to reducing computational costs in convolutional neural networks (CNNs) during inference is presented. The conventional method of passing input data through a fixed architecture is inefficient, as easy examples can be classified at early stages. Early-exit CNNs (EENets) adapt their computational cost based on the input by stopping the inference process at certain exit locations. EENets consist of multiple exit blocks with confidence and softmax branches, which are trained to optimize both classification accuracy and computational cost. During training, the network learns to adapt its many confidence branches to inputs, reducing computation for easy examples. The idea is applicable to available CNN architectures like ResNets. Experimental results on MNIST, SVHN, CIFAR10, and Tiny-ImageNet datasets demonstrate that EENets achieve similar accuracy as non-EENet versions while reducing computational cost by 80%. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper develops a method to make convolutional neural networks (CNNs) more efficient during inference. Right now, CNNs use the same amount of computation for all images, even if some are very easy to classify. The new approach is called Early-exit CNNs or EENets. EENets learn how to stop processing an image early on if it’s already clear what the answer will be. This makes them much faster and uses less energy. The paper shows that EENets can work just as well as regular CNNs, but are 20% faster. |
Keywords
» Artificial intelligence » Classification » Cnn » Inference » Softmax
