en-es
en
0.25
0.5
0.75
1.25
1.5
1.75
2
Surrogate Regret Bounds for the Area Under the ROC Curve via Strongly Proper Losses
Published on Aug 09, 20133816 Views
The area under the ROC curve (AUC) is a widely used performance measure in machine learning, and has been widely studied in recent years particularly in the context of bipartite ranking. A dominant th
Related categories
Chapter list
Surrogate Regret Bounds for the Area Under the ROC Curve via Strongly Proper Losses00:00
Standard Logistic Regression, AdaBoost, and Least Squares Regression are AUC-Consistent!00:07
Area Under the ROC Curve (AUC) - 100:25
Area Under the ROC Curve (AUC) - 200:43
Area Under the ROC Curve (AUC) - 300:53
Area Under the ROC Curve (AUC) - 401:09
Area Under the ROC Curve (AUC) - 501:15
Empirical AUC as a Wilcoxon-Mann-Whitney Statistic01:37
Pairwise Surrogate Risk Minimization Algorithms for Optimizing AUC01:52
Many of these Pairwise Algorithms are AUC-Consistent02:16
Good AUC Performance03:07
This Paper03:44
Road Map - 104:10
Problem Setup04:31
Binary Loss Functions05:34
Example: 0-1 Loss06:12
Example: Logistic Loss06:26
Example: Exponential Loss06:34
Reduction to Pairwise Binary Classification - 106:51
Reduction to Pairwise Binary Classification - 208:32
Regret Bounds via Balanced Losses - 109:37
Regret Bounds via Balanced Losses - 210:30
Summary So Far10:41
Road Map - 212:01
Proper Losses - 112:14
Proper Losses - 213:31
Strongly Proper Losses 14:52
Road Map - 316:13
Regret Bound via Strongly Proper (Composite) Losses16:15
Examples of Strongly Proper Composite Losses18:00
Tighter Bound under Low-Noise Conditions18:10
AUC-consistent!18:38
Strongly proper losses may also be useful in other contexts19:34