The vips4/i-split repository contains the official implementation of the paper “I-SPLIT: Deep Network Interpretability for Split Computing”, which was accepted at ICPR 2022. Here are the key details about this project:
Purpose: The framework aims to identify the most suitable splitting points for deep networks in split computing scenarios, predicting how well a split will perform regarding classification accuracy before implementation.
Methodology: It utilizes deep network interpretability to find optimal, efficient split points for distributed tasks, considering how specific classification classes affect the best splitting choice.
Experiments: The authors conducted tests on popular networks and datasets, including: Networks: VGG16 and ResNet-50.
Datasets: Tiny-Imagenet-200, notMNIST, and Chest X-Ray Pneumonia.
Paper Reference: The underlying research is titled “I-SPLIT: Deep Network Interpretability for Split Computing” and is available via arXiv (2022). If you’d like to dive deeper, I can look for: Specific instructions on how to run their experiments. More details on the results of the experiments.
Information on how it compares to other split computing methods. Let me know what you’d like to explore next! I-SPLIT: Deep Network Interpretability for Split Computing