This is a simple split learning framework, which simulates the split learning procedure in one program while ignoring all the network communication stuff.
Moreover, this code is also the experiment code of the IJCAI'23 paper Reducing Communication for Split Learning by Randomized Top-k Sparsification (ijcai.org) Please cite as
@inproceedings{zf2023randtopk,
title = {Reducing Communication for Split Learning by Randomized Top-k Sparsification},
author = {Zheng, Fei and Chen, Chaochao and Lyu, Lingjuan and Yao, Binhui},
booktitle = {Proceedings of the Thirty-Second International Joint Conference on
Artificial Intelligence, {IJCAI-23}},
publisher = {International Joint Conferences on Artificial Intelligence Organization},
editor = {Edith Elkind},
pages = {4665--4673},
year = {2023},
month = {8},
note = {Main Track},
doi = {10.24963/ijcai.2023/519},
url = {https://doi.org/10.24963/ijcai.2023/519},
}
The experiment codes are in examples/task/ directory.
Notice that the codes in examples/sparse/ are legacy scripts and contain errors, just ignore them!