Technical perspective: Why don't today's deep nets overfit to their training data?
Sanjeev AroraPublished in: Commun. ACM (2021)
Keyphrases
- training data
- training and test data
- test data
- supervised learning
- decision trees
- data sets
- training examples
- learning algorithm
- training set
- test set
- world wide
- database
- support vector machine
- noisy data
- training process
- semi supervised learning
- training instances
- prior knowledge
- training samples
- learned from training data
- unlabeled data
- classification accuracy
- class labels
- training dataset
- feature values
- deep learning
- information retrieval
- real time