BalDRO: A Distributionally Robust Optimization based Framework for Large Language Model Unlearning

TL;DR

BalDRO introduces a distributionally robust framework for large language model unlearning, effectively addressing sample imbalance issues to improve forgetting accuracy and model utility.

Contribution

It proposes a novel BalDRO framework with two variants, BalDRO-G and BalDRO-DV, for balanced and efficient LLM unlearning, advancing beyond existing methods.

Findings

BalDRO significantly enhances unlearning effectiveness.

BalDRO maintains higher model utility after unlearning.

Experiments on TOFU and MUSE datasets validate its superiority.

Abstract

As Large Language Models (LLMs) increasingly shape online content, removing targeted information from well-trained LLMs (also known as LLM unlearning) has become critical for web governance. A key challenge lies in sample-wise imbalance within the forget set: different samples exhibit widely varying unlearning difficulty, leading to asynchronous forgetting where some knowledge remains insufficiently erased while others become over-forgotten. To address this, we propose BalDRO, a novel and efficient framework for balanced LLM unlearning. BalDRO formulates unlearning as a min-sup process: an inner step identifies a worst-case data distribution that emphasizes hard-to-unlearn samples, while an outer step updates model parameters under this distribution. We instantiate BalDRO via two efficient variants: BalDRO-G, a discrete GroupDRO-based approximation focusing on high-loss subsets, and BalDRO-DV, a continuous Donsker-Varadhan dual method enabling smooth adaptive weighting within standard training pipelines. Experiments on TOFU and MUSE show that BalDRO significantly improves both forgetting quality and model utility over existing methods, and we release code for reproducibility.