I have a large dataset(public available) of text that is labelled. However the test distribution (actual production setting of company) while similar is not from the same source and thus tends to fail on some cases. It is not possible to label our test distribution due to manpower issues. How do I incrementally train my Bert Model to better handle failure cases without overfitting? (i.e. we only label failure cases due to complains from clients which are not many.)



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.