Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

CINIC

Benchmarks

Task NameDataset NameSOTA ResultTrend
Image ClassificationCINIC-10 (test)
Accuracy95.8
177
Image ClassificationCINIC-10
Accuracy65
59
Image ClassificationCINIC-10 1.0 (10% label)
Accuracy81.29
42
Image ClassificationCINIC-10 Dir(0.01), 50 clients, 20% participation
Accuracy56.96
36
Image ClassificationCINIC-10 Dir(0.5) (test)
Accuracy62.56
28
Image ClassificationCINIC-10 Dir(0.5)
Accuracy62.96
28
Image ClassificationCINIC-10 under sudden drift and Dir(0.1) 1.0 (test)
Generalized Accuracy46.02
27
Image ClassificationCINIC-10 non-iid
Accuracy37.59
26
Image ClassificationCINIC-10 iid (test)
Test Accuracy40.8
26
Image ClassificationCINIC-10 20 clients (test)
Accuracy51.12
14
Poisoning Defense in U-shape Split LearningCINIC-10
Accuracy67
10
Predicting GeneralizationCINIC FCN PGDL tasks (train test)
CMI Score16.93
10
Membership Inference AttackCINIC-10
AUC0.8166
8
Attack Robustness AnalysisCINIC-10
AUC81.67
8
Membership InferenceCINIC-10
Delta AUC0.081
8
Image ClassificationCINIC10
Communication Rounds5
7
Online Label ShiftCINIC10 Bernoulli shift
Average Error25.63
7
Online Label ShiftCINIC10 Linear shift
Average Error26.21
7
Online Label Shift AdaptationCINIC10 Square shift (test)
Average Error25.56
7
Federated Learning EfficiencyCINIC-10 (test)
Communication Rounds104
6
Poison DefenseCINIC-10 (test)
Avg Poison Success4.76
6
Membership Inference AttackCINIC-10
TPR @ 0.01% FPR0.31
5
Membership Inference AttackCINIC-10
Loss0.2688
4
VFL Attack PerformanceCINIC-10
LISR17.82
4
Generalization Gap PredictionCINIC-10
CMI Score33.76
4
Showing 25 of 28 rows