On the Connection between Pre-training Data Diversity and Fine-tuning Robustness.
Vivek RamanujanThao NguyenSewoong OhLudwig SchmidtAli FarhadiPublished in: CoRR (2023)
Keyphrases
- fine tuning
- training data
- viable alternative
- manually labeled training data
- fine tuned
- fine tune
- training set
- learning algorithm
- training examples
- decision trees
- test data
- data sets
- general purpose
- domain knowledge
- prior knowledge
- supervised learning
- support vector machine
- training samples
- test set
- computational efficiency
- learned from training data
- high robustness
- real time
- labelled data
- neural network
- machine learning
- knowledge base
- unseen data
- feature space
- training process
- generalization error