Topology-Independent Robustness of the Weighted Mean under Label Poisoning Attacks in Heterogeneous Decentralized Learning

Install

1. Download the dependant packages:

• python 3.9.18
• pytorch 1.13.1
• matplotlib 3.5.0
• networkx 2.6.3

2. Download the dataset to the directory ./dataset and create a directory named ./record. The experiment outputs will be stored in ./record.

• MNIST: http://yann.lecun.com/exdb/mnist/
• CIFAR100: https://www.cs.toronto.edu/~kriz/cifar.html

Construction

The main programs can be found in the following files:

• ByrdLab: main codes
• main DSGD(-xxx).py, : program entry
  • main DSGD.py: compute classification accuracies of different aggregators (Fig. 1, 3, 5, 6)
  • main DSGD-hetero-disturb.py: compute heterogeneity of regular gradients and disturbances of poisoned gradients (Fig. 4)
• draw_decentralized_multi_fig: directories containing the codes that draw the figures in paper

Runing

Run DSGD

python "main DSGD.py"  --aggregation <aggregation-name> --attack <attack-name> --data-partition <data-partition> --graph <graph-name> --gpu <gpu-id> 
# ========================
# e.g.
# python "main DSGD.py" --aggregation mean --attack label_flipping --data-partition noniid --graph TwoCastle --gpu 0

The arguments can be

<aggregation-name>:

• mean
• trimmed-mean
• faba
• cc
• scc
• ios
• lfighter

<attack-name>:

• label_flipping (which executes static label flipping attacks)
• dynamic_label_flipping (which executes dynamic label flipping attacks)

<data-partition>:

• iid
• dirichlet_mild
• noniid

<graph-name>:

• TwoCastle
• UnconnectedRegularLine
• Fan
• Lollipop

---

====================

Figure 1

python run-toyexample.py

cd draw_decentralized_multi_fig

python draw-Decentral-toyexample.py

====================

Figure 10

cd draw_decentralized_multi_fig

python draw-Decentral-topology.py

====================

Figures 3, 6, 8

python run-experiments.py

cd draw_decentralized_multi_fig

python draw-Decentral-MultiFig.py

====================

Figure 4

python run-hetero-disturb.py

cd draw_decentralized_multi_fig

python draw-Decentral-A-xi.py