Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks.
Curtis G. NorthcuttAnish AthalyeJonas MuellerPublished in: NeurIPS Datasets and Benchmarks (2021)
Keyphrases
- test set
- machine learning
- error rate
- training set
- test data
- machine learning methods
- training data
- evaluation methodology
- pattern recognition
- data analysis
- artificial intelligence
- multi label
- test cases
- knowledge acquisition
- learning tasks
- machine learning algorithms
- computer vision
- label noise
- explanation based learning
- support vector machine
- computing environments
- active learning
- computer science
- reinforcement learning
- feature selection
- pervasive computing
- supervised learning
- semi supervised learning
- database
- prior knowledge
- learning algorithm
- data mining