Model Collapse Is Not a Bug but a Feature in Machine Unlearning for LLMs
This addresses privacy concerns in LLMs by offering a more comprehensive unlearning approach that aligns with real-world constraints, though it appears incremental as it builds on existing observations about model collapse.
The paper tackles the problem of machine unlearning in LLMs by proposing Partial Model Collapse (PMC), a method that deliberately triggers model collapse to remove private information without using unlearning targets in the objective, and empirically shows it effectively removes private data while preserving model utility.
Current unlearning methods for LLMs optimize on the private information they seek to remove by incorporating it into their fine-tuning data. We argue this not only risks reinforcing exposure to sensitive data, it also fundamentally contradicts the principle of minimizing its use. As a remedy, we propose a novel unlearning method-Partial Model Collapse (PMC), which does not require unlearning targets in the unlearning objective. Our approach is inspired by recent observations that training generative models on their own generations leads to distribution collapse, effectively removing information from model outputs. Our central insight is that model collapse can be leveraged for machine unlearning by deliberately triggering it for data we aim to remove. We theoretically analyze that our approach converges to the desired outcome, i.e. the model unlearns the data targeted for removal. We empirically demonstrate that PMC overcomes three key limitations of existing unlearning methods that explicitly optimize on unlearning targets, and more effectively removes private information from model outputs while preserving general model utility. Overall, our contributions represent an important step toward more comprehensive unlearning that aligns with real-world privacy constraints. Code available at https://www.cs.cit.tum.de/daml/partial-model-collapse/.