Computer Science
2 years ago

Bayes classifier fails in noisy label problems, hindering accurate classification.

arXiv (Cornell University)
Bayes classifier cannot be learned from noisy responses with unknown noise rates
Soham Bakshi, Subha Maity

Paper Summary

Training a classifier with noisy labels is challenging when we don't know the distribution of label noise. The Bayes decision rule is often unidentified in such cases, making it hard to learn the optimal classifier. However, in some special situations where the Bayes decision rule is identified, a simple algorithm has been developed to learn it without needing to know the noise distribution.

Related papers

Theme background image
Mathematics
8 years ago

Stochastic Estimation Revolutionizes Binomial Parameter Estimation in Costly Observations.

Multicollinearity
Model-Based Clustering
1 view
Theme background image
Chemistry
8 years ago

New classifier accurately identifies components in complex systems using masked data.

Chemometrics
Dimensionality Reduction
Multicollinearity
2 views
Theme background image
Mathematics
8 years ago

New Pattern Recognition Method Minimizes Misclassification Risk, Improving Decision Making

Multicollinearity