On the Connection between Pre-training Data Diversity and Fine-tuning Robustness.
Vivek RamanujanThao NguyenSewoong OhAli FarhadiLudwig SchmidtPublished in: NeurIPS (2023)
Keyphrases
- fine tuning
- training data
- viable alternative
- fine tune
- manually labeled training data
- fine tuned
- training set
- learning algorithm
- data sets
- test data
- training examples
- decision trees
- prior knowledge
- computational efficiency
- supervised learning
- training instances
- test set
- artificial intelligence
- data mining
- semi supervised learning
- domain knowledge
- case study
- search engine