Pith. sign in

Atyaephyra at SemEval-2025 Task 4: Low-Rank Negative Preference Optimization

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it
abstract

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.

citation-role summary

background 1

citation-polarity summary

fields

cs.CL 1

years

2025 1

verdicts

CONDITIONAL 1

roles

background 1

polarities

unclear 1

representative citing papers

Learning-Time Encoding Shapes Unlearning in LLMs

cs.CL · 2025-06-18 · conditional · novelty 6.0

How knowledge is encoded during LLM fine-tuning strongly affects later unlearning: paraphrased training data helps unlearning, while entangled chunks hinder selective forgetting.

citing papers explorer

Showing 1 of 1 citing paper.

  • Learning-Time Encoding Shapes Unlearning in LLMs cs.CL · 2025-06-18 · conditional · none · ref 37 · internal anchor

    How knowledge is encoded during LLM fine-tuning strongly affects later unlearning: paraphrased training data helps unlearning, while entangled chunks hinder selective forgetting.