Teaching Natural Language Understanding Models to Generalize Better
Researchers: Dr. Omer Levy (Computer Science), Prof. Jonathan Berant (Computer Science)
Researchers: Dr. Omer Levy (Computer Science), Prof. Jonathan Berant (Computer Science)
Current NLP models can learn many tasks from examples, but fail to generalize to minor deviations from the training set's distribution.
Our goal is to build challenging datasets that focus on generalization, enabling the entire research community to work on this problem.
We also wish to design new models and pretraining methods that improve generalization.