/Sub-FedAvg

Personalized Federated Learning by Structured and Unstructured Pruning under Data Heterogeneity

Primary LanguagePython

Sub-FedAvg

Personalized Federated Learning by Structured and Unstructured Pruning under Data Heterogeneity

This repository contains the pytorch official implementation for the following paper
Personalized Federated Learning by Structured and Unstructured Pruning under Data Heterogeneity
Saeed Vahidian*, Mahdi Morafah*, and Bill Lin
41th IEEE International Conference on Distributed Computing Systems (Won ICDCS Conference Award) (*equal contribution)
YouTube Presentation

If you find our repository and paper useful, please cite our work:

@article{vahidian2021personalized,
  title={Personalized Federated Learning by Structured and Unstructured Pruning under Data Heterogeneity},
  author={Vahidian, Saeed and Morafah, Mahdi and Lin, Bill},
  journal={arXiv preprint arXiv:2105.00562},
  year={2021}
}

Usage

  1. Glone the repository
git clone https://github.com/MMorafah/Sub-FedAvg.git
  1. For Running
  • Sub-FedAvg (Hybrid)
sh script_s.sh
  • Sub-FedAvg (Unstructured)
sh script_u.sh

Dependencies

torch v0.3.1, torchvision v0.2.0

Options

1.General federated options

parser.add_argument('--rounds', type=int, default=300, help="rounds of training")
parser.add_argument('--num_users', type=int, default=100, help="number of users: K")
parser.add_argument('--frac', type=float, default=0.1, help="the fraction of clients: C")
parser.add_argument('--local_ep', type=int, default=5, help="the number of local epochs: E")
parser.add_argument('--local_bs', type=int, default=10, help="local batch size: B")
parser.add_argument('--bs', type=int, default=128, help="test batch size")
parser.add_argument('--lr', type=float, default=0.01, help="learning rate")
parser.add_argument('--momentum', type=float, default=0.5, help="SGD momentum (default: 0.5)")
parser.add_argument('--warmup_epoch', type=int, default=0, help="the number of pretrain local ep")
  1. Model options
# model arguments
parser.add_argument('--model', type=str, default='lenet5', help='model name')
parser.add_argument('--ks', type=int, default=5, help='kernel size to use for convolutions')
parser.add_argument('--in_ch', type=int, default=3, help='input channels of the first conv layer')
  1. dataset partitioning options
# dataset partitioning arguments
parser.add_argument('--dataset', type=str, default='cifar10', 
                    help="name of dataset: mnist, cifar10, cifar100")
parser.add_argument('--nclass', type=int, default=2, help="classes or shards per user")
parser.add_argument('--nsample_pc', type=int, default=250, 
                    help="number of samples per class or shard for each client")
parser.add_argument('--noniid', action='store_true', help='whether i.i.d or not')
parser.add_argument('--shard', action='store_true', help='whether non-i.i.d based on shard or not')
parser.add_argument('--label', action='store_true', help='whether non-i.i.d based on label or not')
parser.add_argument('--split_test', action='store_true', 
                    help='whether split test set in partitioning or not')
  1. Structured (Hybrid) pruning options (main_s.py)
# pruning arguments 
parser.add_argument('--pruning_percent_ch', type=float, default=0.45, 
                    help="Pruning percent for channels (0-1)")
parser.add_argument('--pruning_percent_fc', type=float, default=10, 
                    help="Pruning percent for fully connected layers (0-100)")
parser.add_argument('--pruning_target', type=int, default=90, 
                    help="Total Pruning target percentage (0-100)")
parser.add_argument('--dist_thresh_ch', type=float, default=0.01, 
                    help="threshold for channels masks difference ")
parser.add_argument('--dist_thresh_fc', type=float, default=0.0005, 
                    help="threshold for fcs masks difference ")
parser.add_argument('--acc_thresh', type=int, default=50, 
                    help="accuracy threshold to apply the derived pruning mask")

parser.add_argument('--sparsity-regularization', '-sr', dest='sr', action='store_true',
                     help='train with channel sparsity regularization')
parser.add_argument('--s', type=float, default=0.0001, 
                    help='scale sparse rate (default: 0.0001)')

parser.add_argument('--weight-decay', '--wd', default=1e-4, type=float,
                metavar='W', help='weight decay (default: 1e-4)')
  1. Unstructured pruning options (main_u.py)
parser.add_argument('--pruning_percent', type=float, default=10, 
                        help="Pruning percent for layers (0-100)")
parser.add_argument('--pruning_target', type=int, default=30, 
                  help="Total Pruning target percentage (0-100)")
parser.add_argument('--dist_thresh_fc', type=float, default=0.0001, 
                  help="threshold for fcs masks difference ")
parser.add_argument('--acc_thresh', type=int, default=50, 
                  help="accuracy threshold to apply the derived pruning mask")
parser.add_argument('--weight-decay', '--wd', default=1e-4, type=float,
              metavar='W', help='weight decay (default: 1e-4)')
  1. Other options
# other arguments 
parser.add_argument('--gpu', type=int, default=0, help="GPU ID, -1 for CPU")
parser.add_argument('--is_print', action='store_true', help='verbose print')
parser.add_argument('--print_freq', type=int, default=100, help="printing frequency during training rounds")
parser.add_argument('--seed', type=int, default=1, help='random seed (default: 1)')
parser.add_argument('--load_initial', type=str, default='', help='define initial model path')

Contact