Enabling large language models (LLMs) to unlearn knowledge and capabilities acquired during training has proven vital for ensuring compliance with data regulations and promoting ethical practices in generative AI. Although there are growing interests in developing various unlearning algorithms, it remains unclear how to best formulate the unlearning problem. The most popular formulation uses a weighted sum of forget and retain loss, but it often leads to performance degradation due to the inherent trade-off between forget and retain losses. In this work, we argue that it is important to model the hierarchical structure of the unlearning problem, where the forget problem (which unlearns certain knowledge and/or capabilities) takes priority over the retain problem (which preserves model utility). This hierarchical structure naturally leads to a bi-level optimization formulation where the lower-level objective focuses on minimizing the forget loss, while the upper-level objective aims to maintain the model’s utility. Based on this new formulation, we propose a novel algorithm, termed Bi-Level UnleaRning (), which not only possesses strong theoretical guarantees but more importantly, delivers superior performance. In particular, our extensive experiments demonstrate that consistently outperforms all the state-of-the-art algorithms across various unlearning tasks, models, and metrics.

Unlearning aims to remove copyrighted, sensitive, or private content from large language models (LLMs) without a full retraining. In this work, we develop a multi-task unlearning benchmark LUME that features three tasks: (1) unlearn synthetically generated creative short novels, (2) unlearn synthetic biographies with sensitive information, and (3) unlearn a collection of public biographies. We further release two fine-tuned LLMs of 1B and 7B parameter sizes as the target models. We conduct detailed evaluations of several recently-proposed algorithms and present results on carefully crafted metrics to understand their behavior and limitations.

Machine unlearning has been used to remove unwanted knowledge acquired by large language models (LLMs). In this paper, we examine machine unlearning from an optimization perspective, framing it as a regularized multi-task optimization problem, where one task optimizes a forgetting objective and another optimizes the model performance. In particular, we introduce a normalized gradient difference algorithm, enabling us to have better control over the trade-off between the objectives, while integrating a new, automatic learning rate scheduler. We provide a theoretical analysis and empirically demonstrate the superior performance of among state-of-the-art unlearning methods on the TOFU and MUSE datasets while exhibiting stable training.

We introduce SemEval-2025 Task 4: unlearn- ing sensitive content from Large Language Models (LLMs). The task features 3 subtasks for LLM unlearning spanning different use cases: (1) unlearn long form synthetic creative documents spanning different genres; (2) un- learn short form synthetic biographies contain- ing personally identifiable information (PII), in- cluding fake names, phone number, SSN, email and home addresses, and (3) unlearn real docu- ments sampled from the target model’s training dataset. We received over 100 submissions from over 30 institutions and we summarize the key techniques and lessons in this paper.

Large language models (LLM’s) have been widely used for several applications such as question answering, text classification and clustering. While the preliminary results across the aforementioned tasks looks promising, recent work has dived deep into LLM’s performing poorly for complex Named Entity Recognition (NER) tasks in comparison to fine-tuned pre-trained language models (PLM’s). To enhance wider adoption of LLM’s, our paper investigates the robustness of such LLM NER models and its instruction fine-tuned variants to adversarial attacks. In particular, we propose a novel attack which relies on disentanglement and word attribution techniques where the former aids in learning an embedding capturing both entity and non-entity influences separately, and the latter aids in identifying important words across both components. This is in stark contrast to most techniques which primarily leverage non-entity words for perturbations limiting the space being explored to synthesize effective adversarial examples. Adversarial training results based on our method improves the F1 score over original LLM NER model by 8% and 18% on CoNLL-2003 and Ontonotes 5.0 datasets respectively.