Atyaephyra at SemEval-2025 Task 4: Low-Rank Negative Preference Optimization
This work addresses the challenge of removing sensitive information from large language models, though it appears incremental as it builds on existing optimization techniques.
The authors tackled the problem of unlearning sensitive content from LLMs by applying negative preference optimization with low-rank adaptation, which significantly exceeded shared task baselines.
We present a submission to the SemEval 2025 shared task on unlearning sensitive content from LLMs. Our approach employs negative preference optimization using low-rank adaptation. We show that we can utilize this combination to efficiently compute additional regularization terms, which help with unlearning stabilization. The results of our approach significantly exceed the shared task baselines.