The past a few years have witnessed the great success of large language models, demonstrating powerful capabilities in comprehending textual data and generating human-like languages. Large language models achieve success by being trained on vast amounts of textual data, including online sources with copyrighted content and user-generated knowledge. However, this comes at a cost: the potential risk of exposing users' privacy and violating copyright protections. Thus, to safeguard individuals' "right to be forgotten", there has been increasing interests in machine unlearning -- the process of removing information carried by particular training samples from a model while not deteriorating its predictive quality. This is a challenging task due to the black-box nature of language models. Most existing studies focus on mitigating the impact of those forgot samples upon a model's outputs, and do not explicitly consider the geometric distributions of samples in the latent space of a model. To address this issue, we propose a machine unlearning framework, named Deep Contrastive Unlearning for fine-Tuning (DeepCUT) language models. Our proposed model achieves machine unlearning by directly optimizing the latent space of a model. Comprehensive experiments on real-world datasets demonstrate the effectiveness and efficiency of DeepCUT with consistent and significant improvement over baseline methods.
Recent vision-language models (VLMs) have made remarkable strides in generative modeling with multimodal inputs, particularly text and images. However, their susceptibility to generating harmful content when exposed to unsafe queries raises critical safety concerns. While current alignment strategies primarily rely on supervised safety fine-tuning with curated datasets, we identify a fundamental limitation we call the "safety mirage" where supervised fine-tuning inadvertently reinforces spurious correlations between superficial textual patterns and safety responses, rather than fostering deep, intrinsic mitigation of harm. We show that these spurious correlations leave fine-tuned VLMs vulnerable even to a simple one-word modification-based attack, where substituting a single word in text queries with a spurious correlation-inducing alternative can effectively bypass safeguards. Additionally, these correlations contribute to the over prudence, causing fine-tuned VLMs to refuse benign queries unnecessarily. To address this issue, we show machine unlearning (MU) as a powerful alternative to supervised safety fine-tuning as it avoids biased feature-label mappings and directly removes harmful knowledge from VLMs while preserving their general capabilities. Extensive evaluations across safety benchmarks show that under one-word attacks, MU-based alignment reduces the attack success rate by up to 60.17% and cuts unnecessary rejections by over 84.20%. Codes are available at https://github.com/OPTML-Group/VLM-Safety-MU. WARNING: There exist AI generations that may be offensive in nature.
This study investigates the machine unlearning techniques within the context of large language models (LLMs), referred to as \textit{LLM unlearning}. LLM unlearning offers a principled approach to removing the influence of undesirable data (e.g., sensitive or illegal information) from LLMs, while preserving their overall utility without requiring full retraining. Despite growing research interest, there is no comprehensive survey that systematically organizes existing work and distills key insights; here, we aim to bridge this gap. We begin by introducing the definition and the paradigms of LLM unlearning, followed by a comprehensive taxonomy of existing unlearning studies. Next, we categorize current unlearning approaches, summarizing their strengths and limitations. Additionally, we review evaluation metrics and benchmarks, providing a structured overview of current assessment methodologies. Finally, we outline promising directions for future research, highlighting key challenges and opportunities in the field.
In recent years, Large Language Models (LLMs) have faced increasing demands to selectively remove sensitive information, protect privacy, and comply with copyright regulations through unlearning, by Machine Unlearning. While evaluating unlearning effectiveness is crucial, existing benchmarks are limited in scale and comprehensiveness, typically containing only a few hundred test cases. We identify two critical challenges in generating holistic audit datasets: ensuring audit adequacy and handling knowledge redundancy between forget and retain dataset. To address these challenges, we propose HANKER, an automated framework for holistic audit dataset generation leveraging knowledge graphs to achieve fine-grained coverage and eliminate redundant knowledge. Applying HANKER to the popular MUSE benchmark, we successfully generated over 69,000 and 111,000 audit cases for the News and Books datasets respectively, identifying thousands of knowledge memorization instances that the previous benchmark failed to detect. Our empirical analysis uncovers how knowledge redundancy significantly skews unlearning effectiveness metrics, with redundant instances artificially inflating the observed memorization measurements ROUGE from 19.7% to 26.1% and Entailment Scores from 32.4% to 35.2%, highlighting the necessity of systematic deduplication for accurate assessment.
The rapid growth of artificial intelligence (AI) has raised privacy concerns over user data, leading to regulations like the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). With the essential toolbox provided by machine unlearning, AI service providers are now able to remove user data from their trained models as well as the training datasets, so as to comply with such regulations. However, extensive data redemption can be costly and degrade model accuracy. To balance the cost of unlearning and the privacy protection, we propose a buyer-initiated auction mechanism for data redemption, enabling the service provider to purchase data from willing users with appropriate compensation. This approach does not require the server to have any a priori knowledge about the users' privacy preference, and provides an efficient solution for maximizing the social welfare in the investigated problem.
Unlearnable data (ULD) has emerged as an innovative defense technique to prevent machine learning models from learning meaningful patterns from specific data, thus protecting data privacy and security. By introducing perturbations to the training data, ULD degrades model performance, making it difficult for unauthorized models to extract useful representations. Despite the growing significance of ULD, existing surveys predominantly focus on related fields, such as adversarial attacks and machine unlearning, with little attention given to ULD as an independent area of study. This survey fills that gap by offering a comprehensive review of ULD, examining unlearnable data generation methods, public benchmarks, evaluation metrics, theoretical foundations and practical applications. We compare and contrast different ULD approaches, analyzing their strengths, limitations, and trade-offs related to unlearnability, imperceptibility, efficiency and robustness. Moreover, we discuss key challenges, such as balancing perturbation imperceptibility with model degradation and the computational complexity of ULD generation. Finally, we highlight promising future research directions to advance the effectiveness and applicability of ULD, underscoring its potential to become a crucial tool in the evolving landscape of data protection in machine learning.
Growing data privacy demands, driven by regulations like GDPR and CCPA, require machine unlearning methods capable of swiftly removing the influence of specific training points. Although verified approaches like SISA, using data slicing and checkpointing, achieve efficient unlearning for single models by reverting to intermediate states, these methods struggle in teacher-student knowledge distillation settings. Unlearning in the teacher typically forces costly, complete student retraining due to pervasive information propagation during distillation. Our primary contribution is PURGE (Partitioned Unlearning with Retraining Guarantee for Ensembles), a novel framework integrating verified unlearning with distillation. We introduce constituent mapping and an incremental multi-teacher strategy that partitions the distillation process, confines each teacher constituent's impact to distinct student data subsets, and crucially maintains data isolation. The PURGE framework substantially reduces retraining overhead, requiring only partial student updates when teacher-side unlearning occurs. We provide both theoretical analysis, quantifying significant speed-ups in the unlearning process, and empirical validation on multiple datasets, demonstrating that PURGE achieves these efficiency gains while maintaining student accuracy comparable to standard baselines.
We present LoTUS, a novel Machine Unlearning (MU) method that eliminates the influence of training samples from pre-trained models, avoiding retraining from scratch. LoTUS smooths the prediction probabilities of the model up to an information-theoretic bound, mitigating its over-confidence stemming from data memorization. We evaluate LoTUS on Transformer and ResNet18 models against eight baselines across five public datasets. Beyond established MU benchmarks, we evaluate unlearning on ImageNet1k, a large-scale dataset, where retraining is impractical, simulating real-world conditions. Moreover, we introduce the novel Retrain-Free Jensen-Shannon Divergence (RF-JSD) metric to enable evaluation under real-world conditions. The experimental results show that LoTUS outperforms state-of-the-art methods in terms of both efficiency and effectiveness. Code: https://github.com/cspartalis/LoTUS.
As compliance with privacy regulations becomes increasingly critical, the growing demand for data privacy has highlighted the significance of machine unlearning in many real world applications, such as social network and recommender systems, many of which can be represented as graph-structured data. However, existing graph unlearning algorithms indiscriminately modify edges or nodes from well-trained models without considering the potential impact of such structural modifications on fairness. For example, forgetting links between nodes with different genders in a social network may exacerbate group disparities, leading to significant fairness concerns. To address these challenges, we propose a novel approach that jointly optimizes the graph structure and the corresponding model for fair unlearning tasks. Specifically,our approach rewires the graph to enhance unlearning efficiency by removing redundant edges that hinder forgetting while preserving fairness through targeted edge augmentation. Additionally, we introduce a worst-case evaluation mechanism to assess the reliability of fair unlearning performance. Extensive experiments on real-world datasets demonstrate the effectiveness of the proposed approach in achieving superior unlearning outcomes.
Machine unlearning methods have become increasingly important for selective concept removal in large pre-trained models. While recent work has explored unlearning in Euclidean contrastive vision-language models, the effectiveness of concept removal in hyperbolic spaces remains unexplored. This paper investigates machine unlearning in hyperbolic contrastive learning by adapting Alignment Calibration to MERU, a model that embeds images and text in hyperbolic space to better capture semantic hierarchies. Through systematic experiments and ablation studies, we demonstrate that hyperbolic geometry offers distinct advantages for concept removal, achieving near perfect forgetting with reasonable performance on retained concepts, particularly when scaling to multiple concept removal. Our approach introduces hyperbolic-specific components including entailment calibration and norm regularization that leverage the unique properties of hyperbolic space. Comparative analysis with Euclidean models reveals fundamental differences in unlearning dynamics, with hyperbolic unlearning reorganizing the semantic hierarchy while Euclidean approaches merely disconnect cross-modal associations. These findings not only advance machine unlearning techniques but also provide insights into the geometric properties that influence concept representation and removal in multimodal models. Source code available at https://github.com/alex-pv01/HAC
Model quantization enables efficient deployment of deep neural networks on edge devices through low-bit parameter representation, yet raises critical challenges for implementing machine unlearning (MU) under data privacy regulations. Existing MU methods designed for full-precision models fail to address two fundamental limitations in quantized networks: 1) Noise amplification from label mismatch during data processing, and 2) Gradient imbalance between forgotten and retained data during training. These issues are exacerbated by quantized models' constrained parameter space and discrete optimization. We propose Q-MUL, the first dedicated unlearning framework for quantized models. Our method introduces two key innovations: 1) Similar Labels assignment replaces random labels with semantically consistent alternatives to minimize noise injection, and 2) Adaptive Gradient Reweighting dynamically aligns parameter update contributions from forgotten and retained data. Through systematic analysis of quantized model vulnerabilities, we establish theoretical foundations for these mechanisms. Extensive evaluations on benchmark datasets demonstrate Q-MUL's superiority over existing approaches.
In recent years, Multimodal Large Language Models (MLLMs) have demonstrated remarkable advancements in tasks such as visual question answering, visual understanding, and reasoning. However, this impressive progress relies on vast amounts of data collected from the internet, raising significant concerns about privacy and security. To address these issues, machine unlearning (MU) has emerged as a promising solution, enabling the removal of specific knowledge from an already trained model without requiring retraining from scratch. Although MU for MLLMs has gained attention, current evaluations of its efficacy remain incomplete, and the underlying problem is often poorly defined, which hinders the development of strategies for creating more secure and trustworthy systems. To bridge this gap, we introduce a benchmark, named PEBench, which includes a dataset of personal entities and corresponding general event scenes, designed to comprehensively assess the performance of MU for MLLMs. Through PEBench, we aim to provide a standardized and robust framework to advance research in secure and privacy-preserving multimodal models. We benchmarked 6 MU methods, revealing their strengths and limitations, and shedding light on key challenges and opportunities for MU in MLLMs.
Machine unlearning is an emerging paradigm to remove the influence of specific training data (i.e., the forget set) from a model while preserving its knowledge of the rest of the data (i.e., the retain set). Previous approaches assume the forget data to be uniformly distributed from all training datapoints. However, if the data to unlearn is dominant in one group, we empirically show that performance for this group degrades, leading to fairness issues. This work tackles the overlooked problem of non-uniformly distributed forget sets, which we call group-robust machine unlearning, by presenting a simple, effective strategy that mitigates the performance loss in dominant groups via sample distribution reweighting. Moreover, we present MIU (Mutual Information-aware Machine Unlearning), the first approach for group robustness in approximate machine unlearning. MIU minimizes the mutual information between model features and group information, achieving unlearning while reducing performance degradation in the dominant group of the forget set. Additionally, MIU exploits sample distribution reweighting and mutual information calibration with the original model to preserve group robustness. We conduct experiments on three datasets and show that MIU outperforms standard methods, achieving unlearning without compromising model robustness. Source code available at https://github.com/tdemin16/group-robust_machine_unlearning.
Machine unlearning is a process to remove specific data points from a trained model while maintaining the performance on retain data, addressing privacy or legal requirements. Despite its importance, existing unlearning evaluations tend to focus on logit-based metrics (i.e., accuracy) under small-scale scenarios. We observe that this could lead to a false sense of security in unlearning approaches under real-world scenarios. In this paper, we conduct a new comprehensive evaluation that employs representation-based evaluations of the unlearned model under large-scale scenarios to verify whether the unlearning approaches genuinely eliminate the targeted forget data from the model's representation perspective. Our analysis reveals that current state-of-the-art unlearning approaches either completely degrade the representational quality of the unlearned model or merely modify the classifier (i.e., the last layer), thereby achieving superior logit-based evaluation metrics while maintaining significant representational similarity to the original model. Furthermore, we introduce a rigorous unlearning evaluation setup, in which the forgetting classes exhibit semantic similarity to downstream task classes, necessitating that feature representations diverge significantly from those of the original model, thus enabling a more rigorous evaluation from a representation perspective. We hope our benchmark serves as a standardized protocol for evaluating unlearning algorithms under realistic conditions.
Machine Unlearning has recently garnered significant attention, aiming to selectively remove knowledge associated with specific data while preserving the model's performance on the remaining data. A fundamental challenge in this process is balancing effective unlearning with knowledge retention, as naive optimization of these competing objectives can lead to conflicting gradients, hindering convergence and degrading overall performance. To address this issue, we propose Learning to Unlearn while Retaining, aimed to mitigate gradient conflicts between unlearning and retention objectives. Our approach strategically avoids conflicts through an implicit gradient regularization mechanism that emerges naturally within the proposed framework. This prevents conflicting gradients between unlearning and retention, leading to effective unlearning while preserving the model's utility. We validate our approach across both discriminative and generative tasks, demonstrating its effectiveness in achieving unlearning without compromising performance on remaining data. Our results highlight the advantages of avoiding such gradient conflicts, outperforming existing methods that fail to account for these interactions.
Graph unlearning aims to remove a subset of graph entities (i.e. nodes and edges) from a graph neural network (GNN) trained on the graph. Unlike machine unlearning for models trained on Euclidean-structured data, effectively unlearning a model trained on non-Euclidean-structured data, such as graphs, is challenging because graph entities exhibit mutual dependencies. Existing works utilize graph partitioning, influence function, or additional layers to achieve graph unlearning. However, none of them can achieve high scalability and effectiveness without additional constraints. In this paper, we achieve more effective graph unlearning by utilizing the embedding space. The primary training objective of a GNN is to generate proper embeddings for each node that encapsulates both structural information and node feature representations. Thus, directly optimizing the embedding space can effectively remove the target nodes' information from the model. Based on this intuition, we propose node-level contrastive unlearning (Node-CUL). It removes the influence of the target nodes (unlearning nodes) by contrasting the embeddings of remaining nodes and neighbors of unlearning nodes. Through iterative updates, the embeddings of unlearning nodes gradually become similar to those of unseen nodes, effectively removing the learned information without directly incorporating unseen data. In addition, we introduce a neighborhood reconstruction method that optimizes the embeddings of the neighbors in order to remove influence of unlearning nodes to maintain the utility of the GNN model. Experiments on various graph data and models show that our Node-CUL achieves the best unlearn efficacy and enhanced model utility with requiring comparable computing resources with existing frameworks.
Machine unlearning, where users can request the deletion of a forget dataset, is becoming increasingly important because of numerous privacy regulations. Initial works on ``exact'' unlearning (e.g., retraining) incur large computational overheads. However, while computationally inexpensive, ``approximate'' methods have fallen short of reaching the effectiveness of exact unlearning: models produced fail to obtain comparable accuracy and prediction confidence on both the forget and test (i.e., unseen) dataset. Exploiting this observation, we propose a new unlearning method, Adversarial Machine UNlearning (AMUN), that outperforms prior state-of-the-art (SOTA) methods for image classification. AMUN lowers the confidence of the model on the forget samples by fine-tuning the model on their corresponding adversarial examples. Adversarial examples naturally belong to the distribution imposed by the model on the input space; fine-tuning the model on the adversarial examples closest to the corresponding forget samples (a) localizes the changes to the decision boundary of the model around each forget sample and (b) avoids drastic changes to the global behavior of the model, thereby preserving the model's accuracy on test samples. Using AMUN for unlearning a random $10\%$ of CIFAR-10 samples, we observe that even SOTA membership inference attacks cannot do better than random guessing.
Recent work has shown that diffusion models memorize and reproduce training data examples. At the same time, large copyright lawsuits and legislation such as GDPR have highlighted the need for erasing datapoints from diffusion models. However, retraining from scratch is often too expensive. This motivates the setting of data unlearning, i.e., the study of efficient techniques for unlearning specific datapoints from the training set. Existing concept unlearning techniques require an anchor prompt/class/distribution to guide unlearning, which is not available in the data unlearning setting. General-purpose machine unlearning techniques were found to be either unstable or failed to unlearn data. We therefore propose a family of new loss functions called Subtracted Importance Sampled Scores (SISS) that utilize importance sampling and are the first method to unlearn data with theoretical guarantees. SISS is constructed as a weighted combination between simpler objectives that are responsible for preserving model quality and unlearning the targeted datapoints. When evaluated on CelebA-HQ and MNIST, SISS achieved Pareto optimality along the quality and unlearning strength dimensions. On Stable Diffusion, SISS successfully mitigated memorization on nearly 90% of the prompts we tested.
In this paper, we investigate knowledge forgetting in large language models with a focus on its generalisation--ensuring that models forget not only specific training samples but also related implicit knowledge. To this end, we begin by identifying a broader unlearning scope that includes both target data and logically associated samples, including rephrased, subject-replaced, one-hop reasoned, and relation-reversed data. To rigorously evaluate generalisation, we introduce UGBench, the first comprehensive benchmark specifically designed to assess the unlearning of in-scope implicit knowledge covering 13 state-of-the-art methods across three datasets. UGBench reveals that unlearned models can still recall paraphrased answers and retain target facts in intermediate layers. This motivates us to take a preliminary step toward more generalised implicit knowledge forgetting by proposing PerMU, a novel probability perturbation-based unlearning paradigm. PerMU simulates adversarial unlearning samples to eliminate fact-related tokens from the logit distribution, collectively reducing the probabilities of all answer-associated tokens. Experiments are conducted on a diverse range of datasets, including TOFU, Harry Potter, ZsRE, WMDP, and MUSE, using models ranging from 1.3B to 13B in scale. The results demonstrate that PerMU delivers up to a 50.40% improvement in unlearning vanilla target data while maintaining a 40.73% boost in forgetting implicit knowledge. Our code can be found in https://github.com/MaybeLizzy/UGBench.
Deep learning (DL) enabled semantic communications leverage DL to train encoders and decoders (codecs) to extract and recover semantic information. However, most semantic training datasets contain personal private information. Such concerns call for enormous requirements for specified data erasure from semantic codecs when previous users hope to move their data from the semantic system. {Existing machine unlearning solutions remove data contribution from trained models, yet usually in supervised sole model scenarios. These methods are infeasible in semantic communications that often need to jointly train unsupervised encoders and decoders.} In this paper, we investigate the unlearning problem in DL-enabled semantic communications and propose a semantic communication unlearning (SCU) scheme to tackle the problem. {SCU includes two key components. Firstly,} we customize the joint unlearning method for semantic codecs, including the encoder and decoder, by minimizing mutual information between the learned semantic representation and the erased samples. {Secondly,} to compensate for semantic model utility degradation caused by unlearning, we propose a contrastive compensation method, which considers the erased data as the negative samples and the remaining data as the positive samples to retrain the unlearned semantic models contrastively. Theoretical analysis and extensive experimental results on three representative datasets demonstrate the effectiveness and efficiency of our proposed methods.