Counterfactual Memorization in Neural Language Models

Chiyuan Zhang, Daphne Ippolito, Katherine Lee, Matthew Jagielski, Florian Tramèr, Nicholas Carlini

Keywords: Memorization, Language Models

TL;DR: We study a notion of counterfactual memorization on common language modeling datasets and show that memorized examples exist in all of them and are not easily captured by more standard text matching based notion of memorization.

Abstract:

Modern neural language models that are widely used in various NLP tasks risk memorizing sensitive information from their training data.
Understanding this memorization is important in real world applications and also from a learning-theoretical perspective. An open question in previous studies of language model memorization is how to filter out ``common'' memorization. In fact, most memorization criteria strongly correlate with the number of occurrences in the training set, capturing memorized familiar phrases, public knowledge, templated texts, or other repeated data.
We formulate a notion of counterfactual memorization which characterizes how a model's predictions change if a particular document is omitted during training.
We identify and study counterfactually-memorized training examples in standard text datasets.
We estimate the influence of each memorized training example on the validation set and on generated texts, showing how this can provide direct evidence of the source of memorization at test time.

Submission Number: 5273