List your company
ResearchFormerly WizWand

Malicious Client Detection on Fashion-MNIST

6Avg Malicious Clients DetectedLabel flipping
0.82.153.54.85May 19, 2024
Updated 7mo ago

Evaluation results

MethodLinks
Label flippingM (Number of malicious clients)=6, alpha=30%2024.056——
PoisonGANM (Number of malicious clients)=6, alpha=30%2024.056——
Heavy noiseM (Number of malicious clients)=6, alpha=30%2024.056——
VagueGANM (Number of malicious clients)=6, alpha=30%2024.055.8——
SAP noiseM (Number of malicious clients)=6, alpha=30%2024.055.7——
Light noiseM (Number of malicious clients)=6, alpha=30%2024.055.6——
Label flippingM (Number of malicious clients)=4, alpha=20%2024.054——
PoisonGANM (Number of malicious clients)=4, alpha=20%2024.054——
Heavy noiseM (Number of malicious clients)=4, alpha=20%2024.054——
SAP noiseM (Number of malicious clients)=4, alpha=20%2024.053.9——
VagueGANM (Number of malicious clients)=4, alpha=20%2024.053.9——
Light noiseM (Number of malicious clients)=4, alpha=20%2024.053.8——
Label flippingM (Number of malicious clients)=2, alpha=10%2024.052——
PoisonGANM (Number of malicious clients)=2, alpha=10%2024.052——
Heavy noiseM (Number of malicious clients)=2, alpha=10%2024.052——
SAP noiseM (Number of malicious clients)=2, alpha=10%2024.052——
VagueGANM (Number of malicious clients)=2, alpha=10%2024.052——
Light noiseM (Number of malicious clients)=2, alpha=10%2024.051.9——
Label flippingM (Number of malicious clients)=1, alpha=5%2024.051——
PoisonGANM (Number of malicious clients)=1, alpha=5%2024.051——
Light noiseM (Number of malicious clients)=1, alpha=5%2024.051——
Heavy noiseM (Number of malicious clients)=1, alpha=5%2024.051——
SAP noiseM (Number of malicious clients)=1, alpha=5%2024.051——
VagueGANM (Number of malicious clients)=1, alpha=5%2024.051——
FLANDERSAttack Strategy=GAUSS, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—100100
FLANDERSAttack Strategy=LIE, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—100100
FLANDERSAttack Strategy=OPT, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—100100
FLANDERSAttack Strategy=AGR-MM, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—100100
FLANDERSAttack type=GAUSS, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—11
FLANDERSAttack type=LIE, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—11
FLANDERSAttack type=OPT, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—11
FLANDERSAttack type=AGR-MM, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—11
FLDetectorAttack Strategy=GAUSS, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—8080
FLDetectorAttack Strategy=LIE, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—8080
FLDetectorAttack Strategy=OPT, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—7979
FLDetectorAttack Strategy=AGR-MM, Malicious client ratio (r)=0.8, Training rounds (T)=502023.03—8080
FLDetectorAttack type=GAUSS, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—0.60.6
FLDetectorAttack type=LIE, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—0.590.59
FLDetectorAttack type=OPT, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—0.60.6
FLDetectorAttack type=AGR-MM, Attack ratio (r)=0.6, Number of rounds (T)=502023.03—0.610.61