Calibration with Bias-Corrected Temperature Scaling Improves Domain Adaptation Under Label Shift in Modern Neural Networks.
Avanti ShrikumarAnshul KundajePublished in: CoRR (2019)
Keyphrases
- domain adaptation
- neural network
- semi supervised
- labeled data
- semi supervised learning
- cross domain
- transfer learning
- multiple sources
- unlabeled data
- manifold alignment
- sentiment classification
- class labels
- multi label
- target domain
- source and target domains
- document classification
- label information
- covariate shift
- data sets
- collaborative filtering
- high dimensional
- training data
- source domain
- databases