0.25
0.5
0.75
1.25
1.5
1.75
2
Classification with Asymmetric Label Noise: Consistency and Maximal Denoising
Published on Aug 09, 20133359 Views
In many real-world classification problems, the labels of training examples are randomly corrupted. Thus, the set of training examples for each class is contaminated by examples of the other class. Pr
Related categories
Chapter list
Classification with Asymmetric Label Noise: Consistency and Maximal Denoising00:00
Plan00:16
Standard (generative) setting for classification00:27
Standard classification: general principles01:22
Contamination model01:48
Equivalent model: (asymmetric) random label noise model02:30
Motivating Application04:06
Related work, previous assumptions04:50
Understanding label noise06:33
Label noise under different error criteria - 107:19
Label noise under different error criteria - 207:40
Label noise under different error criteria - 307:56
Label noise under different error criteria - 408:30
We can surely estimate08:43
Only one contaminated distribution - 109:09
Only one contaminated distribution - 209:23
Only one contaminated distribution - 309:30
Only one contaminated distribution - 409:42
Only p0 contaminated: Identifiablity09:53
Only p0 contaminated: estimation10:38
Mutual contamination - 111:36
Mutual contamination - 211:38
Mutual contamination - 311:49
The two representations13:26
Identifiability - 113:51
Identifiability - 214:32
Identifiability - 314:34
Mutual irreducibility - 114:42
Mutual irreducibility - 214:48
Characterizing the irreducible solution16:04
Geometry of solutions16:28
Consistent estimation of contamination proportions16:37
Consistent estimation of risk17:09
Conclusion18:30
Thank you19:30