Deduplicating Training Data Mitigates Privacy Risks in Language Models.

Nikhil Kandpal, Eric Wallace, Colin Raffel

Browse the full ICML paper archive.