Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks.
Curtis G. NorthcuttAnish AthalyeJonas MuellerPublished in: CoRR (2021)
Keyphrases
- test set
- machine learning
- error rate
- training set
- test data
- machine learning algorithms
- pattern recognition
- machine learning methods
- training data
- context aware
- data analysis
- multi label
- feature selection
- learning algorithm
- evaluation methodology
- label noise
- explanation based learning
- knowledge acquisition
- decision trees
- text classification
- natural language processing
- test cases
- class labels
- information extraction
- learning tasks
- data mining
- supervised learning
- knowledge representation
- classification accuracy
- random selection