The official repos. for "MORI-RAN: Multi-view Robust Representation Learning via Hybrid Contrastive Fusion"
-
Submitted at: ICDM 2022 Workshop on Multi-view Representation Learning
-
Status: Accepted
The Caltech101-20, LandUse-21, and Scene-15 datasets are placed in src/data/processed
folder.
The NoisyMNIST dataset could be downloaded from Google Cloud or Baidu Cloud password: 9q5e.
Thanks for COMPLETER.
If want to train the model with fp16
, you need to install the Apex
.
- Step 1: install dependencies.
git clone https://github.com/Guanzhou-Ke/mori-ran.git
cd mori-ran
conda create -n moriran --file requirements.txt
conda activate moriran
- Step 2: run MORIRAN on
NoisyNMIST
cd src
python main.py -d 0
If want to run CONAN
:
python main.py -e conan -d 0
Finally, all experimental results will save at ./src/experiments/results
You can modify the configuration files at ./src/experiments/configs
. We employ the YACS style to write the experimental configs.
- (Apex) The issue of "tuple index out of range" from cached_cast.
Modify the apex/amp/utils.py#cached_cast
as following:
# change this line (line 113)
- if cached_x.grad_fn.next_functions[1][0].variable is not x:
# into this
+ if cached_x.grad_fn.next_functions[0][0].variable is not x:
@INPROCEEDINGS {10031085,
author = {G. Ke and Y. Zhu and Y. Yu},
booktitle = {2022 IEEE International Conference on Data Mining Workshops (ICDMW)},
title = {MORI-RAN: Multi-view Robust Representation Learning via Hybrid Contrastive Fusion},
year = {2022},
volume = {},
issn = {},
pages = {467-474},
doi = {10.1109/ICDMW58026.2022.00068},
url = {https://doi.ieeecomputersociety.org/10.1109/ICDMW58026.2022.00068},
publisher = {IEEE Computer Society},
address = {Los Alamitos, CA, USA},
month = {dec}
}