Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

WinoBias

Benchmarks

Task NameDataset NameSOTA ResultTrend
Fair GenerationWinoBias Race-Pro (extended)
Deviation Ratio0.08
20
Fair GenerationWinoBias Race (standard)
Deviation Ratio0.04
20
Fair GenerationWinoBias Gender-Pro extended
Deviation Ratio0.07
20
Fair GenerationWinoBias Gender (standard)
Deviation Ratio100
20
Out-of-Domain (OOD) Bias EvaluationWinobias
Accuracy0.507
14
Stereotype Fairness IdentificationWinoBias cloze-style (test)
P_stereo43.18
14
Influence EstimationWinoBias (test)
Spearman Correlation0.854
14
Hallucination DetectionWinobias (test)
AUROC54.43
10
Gender Bias in Coreference ResolutionWinoBias
P(Stereo)49.49
7
Coreference ResolutionWinoBias syntax-type-2
ECE0.154
6
Text-to-Image DebiasingWinobias
Librarian Score86
6
Attribute Presence MeasurementWinoBias (Prof)
Attribute Presence94.1
4
Gender Bias MitigationWinoBias 40 occupations
Bias0.146
4
Gender-fair rewritingWinoBias+ (test)
Tokenised WER0.04
4
Image-to-Image EditingWinoBias adapted for I2I editing (test)
Edit Success Rate93.9
3
Fair Image Generation (Race)Extended Winobias Race+ (test)
Analyst77
3
Fair Image Generation (Race)Winobias original (test)
Analyst82
3
Fair Image Generation (Gender)Extended Winobias Gender+ (test)
Analyst Performance54
3
Fair Image Generation (Gender)Winobias original (test)
Analyst70
3
Coreference ResolutionWinoBias (test)
Accuracy85.1
2
Showing 20 of 20 rows