A SHAP-based Active Learning Approach for Creating High-Quality Training Data.
Nailcan KaraYagiz Levent GumeUmit TigrakGokce EzerogluSerdar MolaOmer Burak AkgunArzucan ÖzgürPublished in: Big Data (2022)
Keyphrases
- active learning
- training data
- high quality
- training examples
- training set
- learning algorithm
- annotation effort
- supervised learning
- sample selection
- generalization error
- labeled data
- unlabeled data
- data sets
- test data
- labeled examples
- low quality
- learning strategies
- stratified sampling
- labeling effort
- prior knowledge
- ground truth
- image quality
- learning process
- random sampling
- domain knowledge
- imbalanced data classification
- higher quality
- active learning strategies
- class labels
- training process
- machine learning
- selective sampling
- labeled data for training
- labeled instances
- training samples
- multi class
- support vector machine
- classification accuracy
- decision trees
- learning problems
- transfer learning
- test set
- text categorization
- batch mode
- semi supervised learning
- relevance feedback
- pairwise
- feature selection