This is the official repo of Recognizable Information Bottleneck (RIB).
The code is implemented using PyTorch framework. Use pip
or conda
to install all dependencies:
pip install -r requirements.txt
or
conda install --file requirements.txt
Run the following command to get all the experimental results in the paper:
./run.sh `which python` <your log dir> <your data dir>
It will automatically invoke all the GPUs on the machine to run the experiments. It took about 16 hours to run through on our machine (8 × NVIDIA RTX A4000).
@inproceedings{ijcai2023p0448,
title = {Recognizable Information Bottleneck},
author = {Lyu, Yilin and Liu, Xin and Song, Mingyang and Wang, Xinyue and Peng, Yaxin and Zeng, Tieyong and Jing, Liping},
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 = {4028--4036},
year = {2023},
month = {8},
note = {Main Track},
doi = {10.24963/ijcai.2023/448},
url = {https://doi.org/10.24963/ijcai.2023/448},
}
MIT