Paper: Robust P2P Personalized Learning

Accepted in the 39th International Symposium on Reliable Distributed Systems (SRDS), 2020

Paper: Robust-P2P-Personalized-Learning-SRDS-2020.pdf

Install requirements

pip install -r requirements.txt

Run

python run.py

Experiments

Communication rounds

# MP, confidence False
mp_exp.communication_rounds(**config)
plots.figure(file, config)

Generates a file with name communication_rounds_{number_of_nodes}_N

# MP, confidence True
mp_exp.communication_rounds(**config)
plots.figure(file, config)

Generates a file with name communication_rounds_{number_of_nodes}_C

# CDPL, confidence True
mp_exp.communication_rounds(**config)
plots.figure(file, config)

Generates a file with name communication_rounds_{number_of_nodes}_F

Plot the result using the function plots.plot(number_of_nodes, analysis)

plots.plot(50, "communication_rounds")

Byzantine resilience

Same analogy

# MP/CDPL, confidence True/False
mp_exp.byzantine(**config)

Contribution factor

# CDPL
mp_exp.contribution_factor(**config)
plots.contribution_factor(file)

Byzantine detection precision

# CDPL
file = mp_exp.byzantine_metrics(**config)
plots.byzantine_metrics(file)

data unbalancedness

# MP/CDPL, confidence True/False
mp_exp.data_unbalancedness(**config)

Plot the result using the function plots.plot(50, "data_unbalancedness")

Graph sparsity

# MP/CDPL, confidence True/False
mp_exp.graph_sparsity(**config)

Plot the result using the function plots.plot(50, "graph_sparsity")

Customization

To customize the setting of the prototype edit the parameters in peernet/constants.py

Default params:

DEBUG_LEVEL = 0
START_PORT = 15000
STOP_CONDITION = 100
TEST_SAMPLE = 1000
CF_THRESHOLD = 0.8
EPSILON_FAIRNESS = -0.1
CONFIDENCE_MEASURE = "max"
ACCURACY_METRIC = "accuracy"
...

karimboubouh/peernet