FedPseudo

The implementation of federated survival analysis using FedPseudo framework.

Paper • Installation • Required Packages • Running Experiments • How to Cite • Contact •

Paper

For more details, see full paper FedPseudo: Privacy-Preserving Pseudo Value-Based Deep Learning Models for Federated Survival Analysis.

Installation

From source

Download a local copy of FedPseudo_KDD23 and install from the directory:

git clone https://github.com/umbc-sanjaylab/FedPseudo_KDD23.git
cd FedPseudo_KDD23
pip install .

Configure the environement

conda env create --name FedPseudo
conda activate FedPseudo

Required Packages

• PyTorch
• pycox
• scikit-learn
• ray
• numpy
• pandas

Install the packages

pip install -r requirements.txt

Running Experiments

Here is one example to run this code for centralized training:

for dataset in ["metabric","support", "gbsg"]: 
    for model_name in ["FedPDNN","FedPLSTM","FedPAttn"]:
        for seed in range(1):
            !python main.py --model=$model_name \
                --dataset=$dataset \
                --alg='all_in' \
                --lr=0.001 \
                --batch-size=128 \
                --epochs=1000 \
                --n_parties=1 \
                --comm_round=1 \
                --partition='centralized' \
                --optimizer='adam'\
                --device='cuda:0'\
                --datadir='./Data/'\
                --logdir='./logs/centralized/'$dataset'/' \
                --patience=100\
                --init_seed=$seed\
                --sample=1.0\

Here is one example to run this code for federated training (iid and non-iid):

for part in ["iid", "non-iid"]:
    for dataset in ["metabric","support", "gbsg"]: 
        for model_name in ["FedPDNN","FedPLSTM","FedPAttn"]:
            for seed in range(1):
                !python main.py --model=$model_name \
                    --dataset=$dataset \
                    --alg='fedavg' \
                    --lr=0.001 \
                    --batch-size=128 \
                    --epochs=20 \
                    --n_parties=5 \
                    --comm_round=20 \
                    --partition=$part \
                    --optimizer='adam'\
                    --device='cuda:0'\
                    --datadir='/home/local/AD/mrahman6/0.AAAI_2023/Data/'\
                    --logdir='/home/local/AD/mrahman6/0.AAAI_2023/FedPseudo_Github/logs/'$part'/'$dataset'/' \
                    --patience=20\
                    --sample=1.0\
                    --init_seed=$seed\
                    --sensitivity=2\
                    --epsilon=8.5\
                    # --is_DP\ #Uncomment if DP is enforced

Parameter	Description
model	The model architecture. Options: FedPDNN, FedLSTM, FedPAttn. Default = FedPDNN.
dataset	Dataset to use. Options: metabric, support, gbsg. Default = metabric.
alg	The training algorithm. Options: fedavg. Default = fedavg.
lr	Learning rate for the local models. Default = 0.001.
batch-size	Batch size. Default = 128.
epochs	Number of local training epochs. Default = 20.
n_parties	Number of parties. Default = 5.
comm_round	Number of communication rounds to use. Default = 20.
partition	The partition type. Options: centralized, iid, non-iid. Default = iid
optimizer	Specify the optimizer. Options: adam, sgd, amsgrad. Default = adam.
device	Specify the device to run the program. Default = cuda:0.
datadir	The path of the dataset. Default = ./Data/.
logdir	The path to store the logs. Default = ./logs/.
patience	Number of patience. Default = 20.
init_seed	The initial seed. Default = 0.
sample	Ratio of parties that participate in each communication round. Default = 1.
sensitivity	Sensitivity parameter for differential privacy. Default = 2.0.
epsilon	Privacy budget parameter for differential privacy. Default = 5.5.
is_DP	True if differential privacy is enforced. Default = False.

How to Cite

@inproceedings{rahman2023fedpseudo,
title={FedPseudo: Privacy-Preserving Pseudo Value-Based Deep Learning Models for Federated Survival Analysis},
author={Rahman, Md Mahmudur and Purushotham, Sanjay},
booktitle={Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
pages={1999--2009},
year={2023}
}

Contact

• Md Mahmudur Rahman (mrahman6@umbc.edu)
• Sanjay Purushotham (psanjay@umbc.edu)