To Each (Textual Sequence) Its Own: Improving Memorized-Data Unlearning in Large Language Models
This work addresses privacy and copyright problems in LLMs by improving memorized-data unlearning, though it is incremental as it builds on existing unlearning techniques.
The paper tackled the problem of LLMs memorizing and regurgitating training sequences, which causes privacy and copyright issues, by proposing that each textual sequence should be unlearned differently based on its memorization degree, resulting in new methods that improved unlearning quality and identified best solutions across various model capacities and forget set sizes.
LLMs have been found to memorize training textual sequences and regurgitate verbatim said sequences during text generation time. This fact is known to be the cause of privacy and related (e.g., copyright) problems. Unlearning in LLMs then takes the form of devising new algorithms that will properly deal with these side-effects of memorized data, while not hurting the model's utility. We offer a fresh perspective towards this goal, namely, that each textual sequence to be forgotten should be treated differently when being unlearned based on its degree of memorization within the LLM. We contribute a new metric for measuring unlearning quality, an adversarial attack showing that SOTA algorithms lacking this perspective fail for privacy, and two new unlearning methods based on Gradient Ascent and Task Arithmetic, respectively. A comprehensive performance evaluation across an extensive suite of NLP tasks then mapped the solution space, identifying the best solutions under different scales in model capacities and forget set sizes and quantified the gains of the new approaches.