truenobility303/f2ba

F2BA

Codes are adapted from BOME, which is published in NeurIPS 2022.

Single-Machine Data Hyper-Cleaning

To reproduce our experiment results, first generate the data by running

python -u data_cleaning.py --pretrain 0

Then run

cd ./single_machine
python -u Run_MNIST.py
python -u Run_Fashion.py

To plot the figure, further run

python -u plot2fig.py

Distributed Learnnnable Regularization

To reproduce our experiment results, please download the dataset l2reg.pt at this link and put it under the data folder specified in args.data_path in distributed/l2reg.py.

You can also generate l2reg.pt by setting args.pretrain=True in distributed/l2reg.py.

Then run

cd ./distributed
python -u l2reg.py --alg F2BA
python -u l2reg.py --alg AID
python -u l2reg.py --alg ITD

And you can find the training logs in the folder args.log.

To plot the figure, further run

python -u plot2fig.py

Note that our training logs are also stored as arrays in the file plot2fig.py. You should change it to your own logs to plot the figure.