I-Split: Deep Network Interpretability for Split Computing

The official PyTorch implementation discussed in "I-Split: Deep Network Interpretability for Split Computing" paper. Accepted at the 26TH International Conference on Pattern Recognition 2022 (ICPR2022).

Abstract

This work makes a substantial step in the field of split computing, i.e., how to split a deep neural network to host its early part on an embedded device and the rest on a server. So far, potential split locations have been identified exploiting uniquely architectural aspects, i.e., based on the layer sizes. Here we show that not only the architecture of the layers does matter, but the importance of the neurons contained therein too. It follows that a split should be applied right after a layer with a high density of important neurons, in order to preserve the information flowing until then. Upon this idea, we propose Interpretable Split (I-Split): a procedure that identifies the most suitable splitting points by providing a reliable prediction on how well this split will perform in terms of classification accuracy, beforehand of its effective implementation. As a further major contribution of I-Split, we show that the best choice for the splitting point on a multiclass categorization problem depends also on which specific classes the network has to deal with.

Installation

1. Clone this repo, and we'll call the directory that you cloned as ${ISPLIT_ROOT}.

2. Install dependencies. We use Python >= 3.8 and PyTorch >= 1.7.0.

3. To use the code from sources in a clean way, it is preferable that you set up a conda enviroment, say:

   $ conda create -n isplit python=3.8
   $ conda activate isplit
   $ conda install pytorch==1.7.0 torchvision==0.8.0 cudatoolkit=11.3 -c pytorch
   $ pip install -r requirements.txt
   

Results of I-Split

Candidate splitting points identified by CDE and the CUI curve of I-Split on the VGG16. The star markers show the extra points that our method is able to identify, and not identified by the CDE approach.

Authors

Federico Cunico, Luigi Capogrosso, Francesco Setti, Damiano Carra, Franco Fummi, Marco Cristani

Department of Computer Science, University of Verona, Italy

[email protected]

Citation

If you use I-Split, please, cite the following paper:

@inproceedings{cunico2022split,
  title={I-SPLIT: Deep Network Interpretability for Split Computing},
  author={Cunico, Federico and Capogrosso, Luigi and Setti, Francesco and Carra, Damiano and Fummi, Franco and Cristani, Marco},
  booktitle={2022 26th International Conference on Pattern Recognition (ICPR)},
  pages={2575--2581},
  year={2022},
  organization={IEEE}
}