Unilogit: Robust Machine Unlearning for LLMs Using Uniform-Target Self-Distillation

This paper introduces Unilogit, a novel self-distillation method for machine unlearning in Large Language Models. Unilogit addresses the challenge of selectively forgetting specific information while maintaining overall model utility, a critical task in compliance with data privacy regulations like GDPR. Unlike prior methods that rely on static hyperparameters or starting model outputs, Unilogit dynamically adjusts target logits to achieve a uniform probability for the target token, leveraging the current model's outputs for more accurate self-distillation targets. This approach not only eliminates the need for additional hyperparameters but also enhances the model's ability to approximate the golden targets. Extensive experiments on public benchmarks and an in-house e-commerce dataset demonstrate Unilogit's superior performance in balancing forget and retain objectives, outperforming state-of-the-art methods such as NPO and UnDIAL. Our analysis further reveals Unilogit's robustness across various scenarios, highlighting its practical applicability and effectiveness in achieving efficacious machine unlearning.

OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models

Large language models (LLMs) trained over extensive corpora risk memorizing sensitive, copyrighted, or toxic content. To address this, we propose OBLIVIATE, a robust unlearning framework that removes targeted data while preserving model utility. The framework follows a structured process: extracting target tokens, building retain sets, and fine-tuning with a tailored loss function comprising three components -- masking, distillation, and world fact. Using low-rank adapters (LoRA), it ensures efficiency without compromising unlearning quality. We conduct experiments on multiple datasets, including the Harry Potter series, WMDP, and TOFU, using a comprehensive suite of metrics: forget quality (new document-level memorization score), model utility, and fluency. Results demonstrate its effectiveness in resisting membership inference attacks, minimizing the impact on retained data, and maintaining robustness across diverse scenarios.

Memory-Efficient Distributed Unlearning

Machine unlearning considers the removal of the contribution of a set of data points from a trained model. In a distributed setting, where a server orchestrates training using data available at a set of remote users, unlearning is essential to cope with late-detected malicious or corrupted users. Existing distributed unlearning algorithms require the server to store all model updates observed in training, leading to immense storage overhead for preserving the ability to unlearn. In this work we study lossy compression schemes for facilitating distributed server-side unlearning with limited memory footprint. We propose memory-efficient distributed unlearning (MEDU), a hierarchical lossy compression scheme tailored for server-side unlearning, that integrates user sparsification, differential thresholding, and random lattice coding, to substantially reduce memory footprint. We rigorously analyze MEDU, deriving an upper bound on the difference between the desired model that is trained from scratch and the model unlearned from lossy compressed stored updates. Our bound outperforms the state-of-the-art known bounds for non-compressed decentralized server-side unlearning, even when lossy compression is incorporated. We further provide a numerical study, which shows that suited lossy compression can enable distributed unlearning with notably reduced memory footprint at the server while preserving the utility of the unlearned model.

Erased but Not Forgotten: How Backdoors Compromise Concept Erasure

The expansion of large-scale text-to-image diffusion models has raised growing concerns about their potential to generate undesirable or harmful content, ranging from fabricated depictions of public figures to sexually explicit images. To mitigate these risks, prior work has devised machine unlearning techniques that attempt to erase unwanted concepts through fine-tuning. However, in this paper, we introduce a new threat model, Toxic Erasure (ToxE), and demonstrate how recent unlearning algorithms, including those explicitly designed for robustness, can be circumvented through targeted backdoor attacks. The threat is realized by establishing a link between a trigger and the undesired content. Subsequent unlearning attempts fail to erase this link, allowing adversaries to produce harmful content. We instantiate ToxE via two established backdoor attacks: one targeting the text encoder and another manipulating the cross-attention layers. Further, we introduce Deep Intervention Score-based Attack (DISA), a novel, deeper backdoor attack that optimizes the entire U-Net using a score-based objective, improving the attack's persistence across different erasure methods. We evaluate five recent concept erasure methods against our threat model. For celebrity identity erasure, our deep attack circumvents erasure with up to 82% success, averaging 57% across all erasure methods. For explicit content erasure, ToxE attacks can elicit up to 9 times more exposed body parts, with DISA yielding an average increase by a factor of 2.9. These results highlight a critical security gap in current unlearning strategies.

DualOptim: Enhancing Efficacy and Stability in Machine Unlearning with Dual Optimizers

Existing machine unlearning (MU) approaches exhibit significant sensitivity to hyperparameters, requiring meticulous tuning that limits practical deployment. In this work, we first empirically demonstrate the instability and suboptimal performance of existing popular MU methods when deployed in different scenarios. To address this issue, we propose Dual Optimizer (DualOptim), which incorporates adaptive learning rate and decoupled momentum factors. Empirical and theoretical evidence demonstrates that DualOptim contributes to effective and stable unlearning. Through extensive experiments, we show that DualOptim can significantly boost MU efficacy and stability across diverse tasks, including image classification, image generation, and large language models, making it a versatile approach to empower existing MU algorithms.

Verifying Robust Unlearning: Probing Residual Knowledge in Unlearned Models

Machine Unlearning (MUL) is crucial for privacy protection and content regulation, yet recent studies reveal that traces of forgotten information persist in unlearned models, enabling adversaries to resurface removed knowledge. Existing verification methods only confirm whether unlearning was executed, failing to detect such residual information leaks. To address this, we introduce the concept of Robust Unlearning, ensuring models are indistinguishable from retraining and resistant to adversarial recovery. To empirically evaluate whether unlearning techniques meet this security standard, we propose the Unlearning Mapping Attack (UMA), a post-unlearning verification framework that actively probes models for forgotten traces using adversarial queries. Extensive experiments on discriminative and generative tasks show that existing unlearning techniques remain vulnerable, even when passing existing verification metrics. By establishing UMA as a practical verification tool, this study sets a new standard for assessing and enhancing machine unlearning security.

Fairness and Robustness in Machine Unlearning

Machine unlearning poses the challenge of ``how to eliminate the influence of specific data from a pretrained model'' in regard to privacy concerns. While prior research on approximated unlearning has demonstrated accuracy and efficiency in time complexity, we claim that it falls short of achieving exact unlearning, and we are the first to focus on fairness and robustness in machine unlearning algorithms. Our study presents fairness Conjectures for a well-trained model, based on the variance-bias trade-off characteristic, and considers their relevance to robustness. Our Conjectures are supported by experiments conducted on the two most widely used model architectures, ResNet and ViT, demonstrating the correlation between fairness and robustness: \textit{the higher fairness-gap is, the more the model is sensitive and vulnerable}. In addition, our experiments demonstrate the vulnerability of current state-of-the-art approximated unlearning algorithms to adversarial attacks, where their unlearned models suffer a significant drop in accuracy compared to the exact-unlearned models. We claim that our fairness-gap measurement and robustness metric should be used to evaluate the unlearning algorithm. Furthermore, we demonstrate that unlearning in the intermediate and last layers is sufficient and cost-effective for time and memory complexity.

SHA256 at SemEval-2025 Task 4: Selective Amnesia -- Constrained Unlearning for Large Language Models via Knowledge Isolation

Large language models (LLMs) frequently memorize sensitive information during training, posing risks when deploying publicly accessible models. Current machine unlearning methods struggle to selectively remove specific data associations without degrading overall model capabilities. This paper presents our solution to SemEval-2025 Task 4 on targeted unlearning, which introduces a two-stage methodology that combines causal mediation analysis with layer-specific optimization. Through systematic causal tracing experiments on OLMo architectures (1B and 7B parameters), we identify the critical role of the first few transformer layers (layers 0-5) in storing subject-attribute associations within MLP modules. Building on this insight, we develop a constrained optimization approach that freezes upper layers while applying a novel joint loss function to lower layers-simultaneously maximizing forget set loss via output token cross-entropy penalties and minimizing retain set deviation through adaptive regularization. Our method achieves 2nd place in the 1B model track, demonstrating strong task performance while maintaining 88% of baseline MMLU accuracy. These results establish causal-informed layer optimization as a promising paradigm for efficient, precise unlearning in LLMs, offering a significant step forward in addressing data privacy concerns in AI systems.

LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks

Large language model unlearning has become a critical challenge in ensuring safety and controlled model behavior by removing undesired data-model influences from the pretrained model while preserving general utility. Significant recent efforts have been dedicated to developing LLM unlearning benchmarks such as WMDP (Weapons of Mass Destruction Proxy) and MUSE (Machine Unlearning Six-way Evaluation), facilitating standardized unlearning performance assessment and method comparison. Despite their usefulness, we uncover for the first time a novel coreset effect within these benchmarks. Specifically, we find that LLM unlearning achieved with the original (full) forget set can be effectively maintained using a significantly smaller subset (functioning as a "coreset"), e.g., as little as 5% of the forget set, even when selected at random. This suggests that LLM unlearning in these benchmarks can be performed surprisingly easily, even in an extremely low-data regime. We demonstrate that this coreset effect remains strong, regardless of the LLM unlearning method used, such as NPO (Negative Preference Optimization) and RMU (Representation Misdirection Unlearning), the popular ones in these benchmarks. The surprisingly strong coreset effect is also robust across various data selection methods, ranging from random selection to more sophisticated heuristic approaches. We explain the coreset effect in LLM unlearning through a keyword-based perspective, showing that keywords extracted from the forget set alone contribute significantly to unlearning effectiveness and indicating that current unlearning is driven by a compact set of high-impact tokens rather than the entire dataset. We further justify the faithfulness of coreset-unlearned models along additional dimensions, such as mode connectivity and robustness to jailbreaking attacks. Codes are available at https://github.com/OPTML-Group/MU-Coreset.

When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformers

Task arithmetic refers to editing the pre-trained model by adding a weighted sum of task vectors, each of which is the weight update from the pre-trained model to fine-tuned models for certain tasks. This approach recently gained attention as a computationally efficient inference method for model editing, e.g., multi-task learning, forgetting, and out-of-domain generalization capabilities. However, the theoretical understanding of why task vectors can execute various conceptual operations remains limited, due to the highly non-convexity of training Transformer-based models. To the best of our knowledge, this paper provides the first theoretical characterization of the generalization guarantees of task vector methods on nonlinear Transformers. We consider a conceptual learning setting, where each task is a binary classification problem based on a discriminative pattern. We theoretically prove the effectiveness of task addition in simultaneously learning a set of irrelevant or aligned tasks, as well as the success of task negation in unlearning one task from irrelevant or contradictory tasks. Moreover, we prove the proper selection of linear coefficients for task arithmetic to achieve guaranteed generalization to out-of-domain tasks. All of our theoretical results hold for both dense-weight parameters and their low-rank approximations. Although established in a conceptual setting, our theoretical findings were validated on a practical machine unlearning task using the large language model Phi-1.5 (1.3B).