Yihua Zhang

Aochuan Chen, Yuguang Yao, Pin-Yu Chen et al.

We revisit and advance visual prompting (VP), an input prompting technique for vision tasks. VP can reprogram a fixed, pre-trained source model to accomplish downstream tasks in the target domain by simply incorporating universal prompts (in terms of input perturbation patterns) into downstream data points. Yet, it remains elusive why VP stays effective even given a ruleless label mapping (LM) between the source classes and the target classes. Inspired by the above, we ask: How is LM interrelated with VP? And how to exploit such a relationship to improve its accuracy on target tasks? We peer into the influence of LM on VP and provide an affirmative answer that a better 'quality' of LM (assessed by mapping precision and explanation) can consistently improve the effectiveness of VP. This is in contrast to the prior art where the factor of LM was missing. To optimize LM, we propose a new VP framework, termed ILM-VP (iterative label mapping-based visual prompting), which automatically re-maps the source labels to the target labels and progressively improves the target task accuracy of VP. Further, when using a contrastive language-image pretrained (CLIP) model, we propose to integrate an LM process to assist the text prompt selection of CLIP and to improve the target task accuracy. Extensive experiments demonstrate that our proposal significantly outperforms state-of-the-art VP methods. As highlighted below, we show that when reprogramming an ImageNet-pretrained ResNet-18 to 13 target tasks, our method outperforms baselines by a substantial margin, e.g., 7.9% and 6.7% accuracy improvements in transfer learning to the target Flowers102 and CIFAR100 datasets. Besides, our proposal on CLIP-based VP provides 13.7% and 7.1% accuracy improvements on Flowers102 and DTD respectively. Our code is available at https://github.com/OPTML-Group/ILM-VP.

Yihua Zhang, Ruisi Cai, Tianlong Chen et al.

Sparsely-gated Mixture of Expert (MoE), an emerging deep model architecture, has demonstrated a great promise to enable high-accuracy and ultra-efficient model inference. Despite the growing popularity of MoE, little work investigated its potential to advance convolutional neural networks (CNNs), especially in the plane of adversarial robustness. Since the lack of robustness has become one of the main hurdles for CNNs, in this paper we ask: How to adversarially robustify a CNN-based MoE model? Can we robustly train it like an ordinary CNN model? Our pilot study shows that the conventional adversarial training (AT) mechanism (developed for vanilla CNNs) no longer remains effective to robustify an MoE-CNN. To better understand this phenomenon, we dissect the robustness of an MoE-CNN into two dimensions: Robustness of routers (i.e., gating functions to select data-specific experts) and robustness of experts (i.e., the router-guided pathways defined by the subnetworks of the backbone CNN). Our analyses show that routers and experts are hard to adapt to each other in the vanilla AT. Thus, we propose a new router-expert alternating Adversarial training framework for MoE, termed AdvMoE. The effectiveness of our proposal is justified across 4 commonly-used CNN model architectures over 4 benchmark datasets. We find that AdvMoE achieves 1% ~ 4% adversarial robustness improvement over the original dense CNN, and enjoys the efficiency merit of sparsity-gated MoE, leading to more than 50% inference cost reduction. Codes are available at https://github.com/OPTML-Group/Robust-MoE-CNN.

Tianlong Chen, Zhenyu Zhang, Yihua Zhang et al.

Trojan attacks threaten deep neural networks (DNNs) by poisoning them to behave normally on most samples, yet to produce manipulated results for inputs attached with a particular trigger. Several works attempt to detect whether a given DNN has been injected with a specific trigger during the training. In a parallel line of research, the lottery ticket hypothesis reveals the existence of sparse subnetworks which are capable of reaching competitive performance as the dense network after independent training. Connecting these two dots, we investigate the problem of Trojan DNN detection from the brand new lens of sparsity, even when no clean training data is available. Our crucial observation is that the Trojan features are significantly more stable to network pruning than benign features. Leveraging that, we propose a novel Trojan network detection regime: first locating a "winning Trojan lottery ticket" which preserves nearly full Trojan information yet only chance-level performance on clean inputs; then recovering the trigger embedded in this already isolated subnetwork. Extensive experiments on various datasets, i.e., CIFAR-10, CIFAR-100, and ImageNet, with different network architectures, i.e., VGG-16, ResNet-18, ResNet-20s, and DenseNet-100 demonstrate the effectiveness of our proposal. Codes are available at https://github.com/VITA-Group/Backdoor-LTH.

Haomin Zhuang, Yihua Zhang, Sijia Liu

Despite the record-breaking performance in Text-to-Image (T2I) generation by Stable Diffusion, less research attention is paid to its adversarial robustness. In this work, we study the problem of adversarial attack generation for Stable Diffusion and ask if an adversarial text prompt can be obtained even in the absence of end-to-end model queries. We call the resulting problem 'query-free attack generation'. To resolve this problem, we show that the vulnerability of T2I models is rooted in the lack of robustness of text encoders, e.g., the CLIP text encoder used for attacking Stable Diffusion. Based on such insight, we propose both untargeted and targeted query-free attacks, where the former is built on the most influential dimensions in the text embedding space, which we call steerable key dimensions. By leveraging the proposed attacks, we empirically show that only a five-character perturbation to the text prompt is able to cause the significant content shift of synthesized images using Stable Diffusion. Moreover, we show that the proposed target attack can precisely steer the diffusion model to scrub the targeted image content without causing much change in untargeted image content. Our code is available at https://github.com/OPTML-Group/QF-Attack.

Aochuan Chen, Yimeng Zhang, Jinghan Jia et al.

Zeroth-order (ZO) optimization has become a popular technique for solving machine learning (ML) problems when first-order (FO) information is difficult or impossible to obtain. However, the scalability of ZO optimization remains an open problem: Its use has primarily been limited to relatively small-scale ML problems, such as sample-wise adversarial attack generation. To our best knowledge, no prior work has demonstrated the effectiveness of ZO optimization in training deep neural networks (DNNs) without a significant decrease in performance. To overcome this roadblock, we develop DeepZero, a principled ZO deep learning (DL) framework that can scale ZO optimization to DNN training from scratch through three primary innovations. First, we demonstrate the advantages of coordinatewise gradient estimation (CGE) over randomized vector-wise gradient estimation in training accuracy and computational efficiency. Second, we propose a sparsityinduced ZO training protocol that extends the model pruning methodology using only finite differences to explore and exploit the sparse DL prior in CGE. Third, we develop the methods of feature reuse and forward parallelization to advance the practical implementations of ZO training. Our extensive experiments show that DeepZero achieves state-of-the-art (SOTA) accuracy on ResNet-20 trained on CIFAR-10, approaching FO training performance for the first time. Furthermore, we show the practical utility of DeepZero in applications of certified adversarial defense and DL-based partial differential equation error correction, achieving 10-20% improvement over SOTA. We believe our results will inspire future research on scalable ZO optimization and contribute to advancing DL with black box. Codes are available at https://github.com/OPTML-Group/DeepZero.

Chongyu Fan, Jiancheng Liu, Yihua Zhang et al.

With evolving data regulations, machine unlearning (MU) has become an important tool for fostering trust and safety in today's AI models. However, existing MU methods focusing on data and/or weight perspectives often suffer limitations in unlearning accuracy, stability, and cross-domain applicability. To address these challenges, we introduce the concept of 'weight saliency' for MU, drawing parallels with input saliency in model explanation. This innovation directs MU's attention toward specific model weights rather than the entire model, improving effectiveness and efficiency. The resultant method that we call saliency unlearning (SalUn) narrows the performance gap with 'exact' unlearning (model retraining from scratch after removing the forgetting data points). To the best of our knowledge, SalUn is the first principled MU approach that can effectively erase the influence of forgetting data, classes, or concepts in both image classification and generation tasks. As highlighted below, For example, SalUn yields a stability advantage in high-variance random data forgetting, e.g., with a 0.2% gap compared to exact unlearning on the CIFAR-10 dataset. Moreover, in preventing conditional diffusion models from generating harmful images, SalUn achieves nearly 100% unlearning accuracy, outperforming current state-of-the-art baselines like Erased Stable Diffusion and Forget-Me-Not. Codes are available at https://github.com/OPTML-Group/Unlearn-Saliency. (WARNING: This paper contains model outputs that may be offensive in nature.)

Yimeng Zhang, Jinghan Jia, Xin Chen et al.

The recent advances in diffusion models (DMs) have revolutionized the generation of realistic and complex images. However, these models also introduce potential safety hazards, such as producing harmful content and infringing data copyrights. Despite the development of safety-driven unlearning techniques to counteract these challenges, doubts about their efficacy persist. To tackle this issue, we introduce an evaluation framework that leverages adversarial prompts to discern the trustworthiness of these safety-driven DMs after they have undergone the process of unlearning harmful concepts. Specifically, we investigated the adversarial robustness of DMs, assessed by adversarial prompts, when eliminating unwanted concepts, styles, and objects. We develop an effective and efficient adversarial prompt generation approach for DMs, termed UnlearnDiffAtk. This method capitalizes on the intrinsic classification abilities of DMs to simplify the creation of adversarial prompts, thereby eliminating the need for auxiliary classification or diffusion models. Through extensive benchmarking, we evaluate the robustness of widely-used safety-driven unlearned DMs (i.e., DMs after unlearning undesirable concepts, styles, or objects) across a variety of tasks. Our results demonstrate the effectiveness and efficiency merits of UnlearnDiffAtk over the state-of-the-art adversarial prompt generation method and reveal the lack of robustness of current safetydriven unlearning techniques when applied to DMs. Codes are available at https://github.com/OPTML-Group/Diffusion-MU-Attack. WARNING: There exist AI generations that may be offensive in nature.

Gaoyuan Zhang, Songtao Lu, Yihua Zhang et al.

Current deep neural networks (DNNs) are vulnerable to adversarial attacks, where adversarial perturbations to the inputs can change or manipulate classification. To defend against such attacks, an effective and popular approach, known as adversarial training (AT), has been shown to mitigate the negative impact of adversarial attacks by virtue of a min-max robust training method. While effective, it remains unclear whether it can successfully be adapted to the distributed learning context. The power of distributed optimization over multiple machines enables us to scale up robust training over large models and datasets. Spurred by that, we propose distributed adversarial training (DAT), a large-batch adversarial training framework implemented over multiple machines. We show that DAT is general, which supports training over labeled and unlabeled data, multiple types of attack generation methods, and gradient compression operations favored for distributed optimization. Theoretically, we provide, under standard conditions in the optimization theory, the convergence rate of DAT to the first-order stationary points in general non-convex settings. Empirically, we demonstrate that DAT either matches or outperforms state-of-the-art robust accuracies and achieves a graceful training speedup (e.g., on ResNet-50 under ImageNet). Codes are available at https://github.com/dat-2022/dat.

Guanhua Zhang, Yihua Zhang, Yang Zhang et al.

Despite a surge of recent advances in promoting machine Learning (ML) fairness, the existing mainstream approaches mostly require retraining or finetuning the entire weights of the neural network to meet the fairness criteria. However, this is often infeasible in practice for those large-scale trained models due to large computational and storage costs, low data efficiency, and model privacy issues. In this paper, we propose a new generic fairness learning paradigm, called FairReprogram, which incorporates the model reprogramming technique. Specifically, FairReprogram considers the case where models can not be changed and appends to the input a set of perturbations, called the fairness trigger, which is tuned towards the fairness criteria under a min-max formulation. We further introduce an information-theoretic framework that explains why and under what conditions fairness goals can be achieved using the fairness trigger. We show both theoretically and empirically that the fairness trigger can effectively obscure demographic biases in the output prediction of fixed ML models by providing false demographic information that hinders the model from utilizing the correct demographic information to make the prediction. Extensive experiments on both NLP and CV datasets demonstrate that our method can achieve better fairness improvements than retraining-based methods with far less data dependency under two widely-used fairness criteria. Codes are available at https://github.com/UCSB-NLP-Chang/Fairness-Reprogramming.git.

Yihua Zhang, Yimeng Zhang, Aochuan Chen et al.

Massive data is often considered essential for deep learning applications, but it also incurs significant computational and infrastructural costs. Therefore, dataset pruning (DP) has emerged as an effective way to improve data efficiency by identifying and removing redundant training samples without sacrificing performance. In this work, we aim to address the problem of DP for transfer learning, i.e., how to prune a source dataset for improved pretraining efficiency and lossless finetuning accuracy on downstream target tasks. To our best knowledge, the problem of DP for transfer learning remains open, as previous studies have primarily addressed DP and transfer learning as separate problems. By contrast, we establish a unified viewpoint to integrate DP with transfer learning and find that existing DP methods are not suitable for the transfer learning paradigm. We then propose two new DP methods, label mapping and feature mapping, for supervised and self-supervised pretraining settings respectively, by revisiting the DP problem through the lens of source-target domain mapping. Furthermore, we demonstrate the effectiveness of our approach on numerous transfer learning tasks. We show that source data classes can be pruned by up to 40% ~ 80% without sacrificing downstream performance, resulting in a significant 2 ~ 5 times speed-up during the pretraining stage. Besides, our proposal exhibits broad applicability and can improve other computationally intensive transfer learning techniques, such as adversarial pretraining. Codes are available at https://github.com/OPTML-Group/DP4TL.

Yize Li, Yihua Zhang, Sijia Liu et al.

Despite the remarkable generation capabilities of Diffusion Models (DMs), conducting training and inference remains computationally expensive. Previous works have been devoted to accelerating diffusion sampling, but achieving data-efficient diffusion training has often been overlooked. In this work, we investigate efficient diffusion training from the perspective of dataset pruning. Inspired by the principles of data-efficient training for generative models such as generative adversarial networks (GANs), we first extend the data selection scheme used in GANs to DM training, where data features are encoded by a surrogate model, and a score criterion is then applied to select the coreset. To further improve the generation performance, we employ a class-wise reweighting approach, which derives class weights through distributionally robust optimization (DRO) over a pre-trained reference DM. For a pixel-wise DM (DDPM) on CIFAR-10, experiments demonstrate the superiority of our methodology over existing approaches and its effectiveness in image synthesis comparable to that of the original full-data model while achieving the speed-up between 2.34 times and 8.32 times. Additionally, our method could be generalized to latent DMs (LDMs), e.g., Masked Diffusion Transformer (MDT) and Stable Diffusion (SD), and achieves competitive generation capability on ImageNet. Code is available here (https://github.com/Yeez-lee/Data-Selection-and-Reweighting-for-Diffusion-Models).

Yihua Zhang, Yuguang Yao, Parikshit Ram et al.

The deployment constraints in practical applications necessitate the pruning of large-scale deep learning models, i.e., promoting their weight sparsity. As illustrated by the Lottery Ticket Hypothesis (LTH), pruning also has the potential of improving their generalization ability. At the core of LTH, iterative magnitude pruning (IMP) is the predominant pruning method to successfully find 'winning tickets'. Yet, the computation cost of IMP grows prohibitively as the targeted pruning ratio increases. To reduce the computation overhead, various efficient 'one-shot' pruning methods have been developed, but these schemes are usually unable to find winning tickets as good as IMP. This raises the question of how to close the gap between pruning accuracy and pruning efficiency? To tackle it, we pursue the algorithmic advancement of model pruning. Specifically, we formulate the pruning problem from a fresh and novel viewpoint, bi-level optimization (BLO). We show that the BLO interpretation provides a technically-grounded optimization base for an efficient implementation of the pruning-retraining learning paradigm used in IMP. We also show that the proposed bi-level optimization-oriented pruning method (termed BiP) is a special class of BLO problems with a bi-linear problem structure. By leveraging such bi-linearity, we theoretically show that BiP can be solved as easily as first-order optimization, thus inheriting the computation efficiency. Through extensive experiments on both structured and unstructured pruning with 5 model architectures and 4 data sets, we demonstrate that BiP can find better winning tickets than IMP in most cases, and is computationally as efficient as the one-shot pruning schemes, demonstrating 2-7 times speedup over IMP for the same level of model accuracy and sparsity.

Jinghan Jia, Yihua Zhang, Dogyoon Song et al.

Lifelong learning (LL) aims to improve a predictive model as the data source evolves continuously. Most work in this learning paradigm has focused on resolving the problem of 'catastrophic forgetting,' which refers to a notorious dilemma between improving model accuracy over new data and retaining accuracy over previous data. Yet, it is also known that machine learning (ML) models can be vulnerable in the sense that tiny, adversarial input perturbations can deceive the models into producing erroneous predictions. This motivates the research objective of this paper - specification of a new LL framework that can salvage model robustness (against adversarial attacks) from catastrophic forgetting. Specifically, we propose a new memory-replay LL strategy that leverages modern bi-level optimization techniques to determine the 'coreset' of the current data (i.e., a small amount of data to be memorized) for ease of preserving adversarial robustness over time. We term the resulting LL framework 'Data-Efficient Robustness-Preserving LL' (DERPLL). The effectiveness of DERPLL is evaluated for class-incremental image classification using ResNet-18 over the CIFAR-10 dataset. Experimental results show that DERPLL outperforms the conventional coreset-guided LL baseline and achieves a substantial improvement in both standard accuracy and robust accuracy.

Yihua Zhang, Pranay Sharma, Parikshit Ram et al.

Invariant risk minimization (IRM) has received increasing attention as a way to acquire environment-agnostic data representations and predictions, and as a principled solution for preventing spurious correlations from being learned and for improving models' out-of-distribution generalization. Yet, recent works have found that the optimality of the originally-proposed IRM optimization (IRM) may be compromised in practice or could be impossible to achieve in some scenarios. Therefore, a series of advanced IRM algorithms have been developed that show practical improvement over IRM. In this work, we revisit these recent IRM advancements, and identify and resolve three practical limitations in IRM training and evaluation. First, we find that the effect of batch size during training has been chronically overlooked in previous studies, leaving room for further improvement. We propose small-batch training and highlight the improvements over a set of large-batch optimization techniques. Second, we find that improper selection of evaluation environments could give a false sense of invariance for IRM. To alleviate this effect, we leverage diversified test-time environments to precisely characterize the invariance of IRM when applied in practice. Third, we revisit (Ahuja et al. (2020))'s proposal to convert IRM into an ensemble game and identify a limitation when a single invariant predictor is desired instead of an ensemble of individual predictors. We propose a new IRM variant to address this limitation based on a novel viewpoint of ensemble IRM games as consensus-constrained bi-level optimization. Lastly, we conduct extensive experiments (covering 7 existing IRM variants and 7 datasets) to justify the practical significance of revisiting IRM training and evaluation in a principled manner.

Yihua Zhang, Prashant Khanduri, Ioannis Tsaknakis et al.

Recently, bi-level optimization (BLO) has taken center stage in some very exciting developments in the area of signal processing (SP) and machine learning (ML). Roughly speaking, BLO is a classical optimization problem that involves two levels of hierarchy (i.e., upper and lower levels), wherein obtaining the solution to the upper-level problem requires solving the lower-level one. BLO has become popular largely because it is powerful in modeling problems in SP and ML, among others, that involve optimizing nested objective functions. Prominent applications of BLO range from resource allocation for wireless systems to adversarial machine learning. In this work, we focus on a class of tractable BLO problems that often appear in SP and ML applications. We provide an overview of some basic concepts of this class of BLO problems, such as their optimality conditions, standard algorithms (including their optimization principles and practical implementations), as well as how they can be leveraged to obtain state-of-the-art results for a number of key SP and ML applications. Further, we discuss some recent advances in BLO theory, its implications for applications, and point out some limitations of the state-of-the-art that require significant future research efforts. Overall, we hope that this article can serve to accelerate the adoption of BLO as a generic tool to model, analyze, and innovate on a wide array of emerging SP and ML applications.

Bairu Hou, Jinghan Jia, Yihua Zhang et al.

Robustness evaluation against adversarial examples has become increasingly important to unveil the trustworthiness of the prevailing deep models in natural language processing (NLP). However, in contrast to the computer vision domain where the first-order projected gradient descent (PGD) is used as the benchmark approach to generate adversarial examples for robustness evaluation, there lacks a principled first-order gradient-based robustness evaluation framework in NLP. The emerging optimization challenges lie in 1) the discrete nature of textual inputs together with the strong coupling between the perturbation location and the actual content, and 2) the additional constraint that the perturbed text should be fluent and achieve a low perplexity under a language model. These challenges make the development of PGD-like NLP attacks difficult. To bridge the gap, we propose TextGrad, a new attack generator using gradient-driven optimization, supporting high-accuracy and high-quality assessment of adversarial robustness in NLP. Specifically, we address the aforementioned challenges in a unified optimization framework. And we develop an effective convex relaxation method to co-optimize the continuously-relaxed site selection and perturbation variables and leverage an effective sampling method to establish an accurate mapping from the continuous optimization variables to the discrete textual perturbations. Moreover, as a first-order attack generation method, TextGrad can be baked into adversarial training to further improve the robustness of NLP models. Extensive experiments are provided to demonstrate the effectiveness of TextGrad not only in attack generation for robustness evaluation but also in adversarial defense.

Yihua Zhang, Hongkang Li, Yuguang Yao et al.

Visual prompting (VP) has emerged as a popular method to repurpose pretrained vision models for adaptation to downstream tasks. Unlike conventional model fine-tuning techniques, VP introduces a universal perturbation directly into the input data to facilitate task-specific fine-tuning rather than modifying model parameters. However, there exists a noticeable performance gap between VP and conventional fine-tuning methods, highlighting an unexplored realm in theory and practice to understand and advance the input-level VP to reduce its current performance gap. Towards this end, we introduce a generalized concept, termed activation prompt (AP), which extends the scope of the input-level VP by enabling universal perturbations to be applied to activation maps within the intermediate layers of the model. By using AP to revisit the problem of VP and employing it as an analytical tool, we demonstrate the intrinsic limitations of VP in both performance and efficiency, revealing why input-level prompting may lack effectiveness compared to AP, which exhibits a model-dependent layer preference. We show that AP is closely related to normalization tuning in convolutional neural networks and vision transformers, although each model type has distinct layer preferences for prompting. We also theoretically elucidate the rationale behind such a preference by analyzing global features across layers. Through extensive experiments across 29 datasets and various model architectures, we provide a comprehensive performance analysis of AP, comparing it with VP and parameter-efficient fine-tuning baselines. Our results demonstrate AP's superiority in both accuracy and efficiency, considering factors such as time, parameters, memory usage, and throughput.

Yihua Zhang, Pingzhi Li, Junyuan Hong et al.

In the evolving landscape of natural language processing (NLP), fine-tuning pre-trained Large Language Models (LLMs) with first-order (FO) optimizers like SGD and Adam has become standard. Yet, as LLMs grow {in size}, the substantial memory overhead from back-propagation (BP) for FO gradient computation presents a significant challenge. Addressing this issue is crucial, especially for applications like on-device training where memory efficiency is paramount. This paper proposes a shift towards BP-free, zeroth-order (ZO) optimization as a solution for reducing memory costs during LLM fine-tuning, building on the initial concept introduced by MeZO. Unlike traditional ZO-SGD methods, our work expands the exploration to a wider array of ZO optimization techniques, through a comprehensive, first-of-its-kind benchmarking study across five LLM families (Roberta, OPT, LLaMA, Vicuna, Mistral), three task complexities, and five fine-tuning schemes. Our study unveils previously overlooked optimization principles, highlighting the importance of task alignment, the role of the forward gradient method, and the balance between algorithm complexity and fine-tuning performance. We further introduce novel enhancements to ZO optimization, including block-wise descent, hybrid training, and gradient sparsity. Our study offers a promising direction for achieving further memory-efficient LLM fine-tuning. Codes to reproduce all our experiments are at https://github.com/ZO-Bench/ZO-LLM .

Jinghan Jia, Yihua Zhang, Yimeng Zhang et al.

Large Language Models (LLMs) have highlighted the necessity of effective unlearning mechanisms to comply with data regulations and ethical AI practices. LLM unlearning aims at removing undesired data influences and associated model capabilities without compromising utility beyond the scope of unlearning. While interest in studying LLM unlearning is growing, the impact of the optimizer choice for LLM unlearning remains unexplored. In this work, we shed light on the significance of optimizer selection in LLM unlearning for the first time, establishing a clear connection between second-order optimization and influence unlearning (a classical approach using influence functions to update the model for data influence removal). This insight propels us to develop a second-order optimization-based LLM unlearning framework, termed Second-Order UnLearning (SOUL), which extends the static, one-shot model update using influence unlearning to a dynamic, iterative unlearning process. Our extensive experiments show that SOUL consistently outperforms conventional first-order methods across various unlearning tasks, models, and metrics, indicating that second-order optimization offers an effective and broadly applicable solution for LLM unlearning. Codes are available at https://github.com/OPTML-Group/SOUL.

Chongyu Fan, Jinghan Jia, Yihua Zhang et al.

The LLM unlearning technique has recently been introduced to comply with data regulations and address the safety and ethical concerns of LLMs by removing the undesired data-model influence. However, state-of-the-art unlearning methods face a critical vulnerability: they are susceptible to ``relearning'' the removed information from a small number of forget data points, known as relearning attacks. In this paper, we systematically investigate how to make unlearned models robust against such attacks. For the first time, we establish a connection between robust unlearning and sharpness-aware minimization (SAM) through a unified robust optimization framework, in an analogy to adversarial training designed to defend against adversarial attacks. Our analysis for SAM reveals that smoothness optimization plays a pivotal role in mitigating relearning attacks. Thus, we further explore diverse smoothing strategies to enhance unlearning robustness. Extensive experiments on benchmark datasets, including WMDP and MUSE, demonstrate that SAM and other smoothness optimization approaches consistently improve the resistance of LLM unlearning to relearning attacks. Notably, smoothness-enhanced unlearning also helps defend against (input-level) jailbreaking attacks, broadening our proposal's impact in robustifying LLM unlearning. Codes are available at https://github.com/OPTML-Group/Unlearn-Smooth.

Yicheng Lang, Changsheng Wang, Yihua Zhang et al.

Zeroth-order (ZO) optimization provides a gradient-free alternative to first-order (FO) methods by estimating gradients via finite differences of function evaluations, and has recently emerged as a memory-efficient paradigm for fine-tuning large-scale models by avoiding backpropagation. However, ZO optimization has a fundamental tension between accuracy and query efficiency. In this work, we show that ZO optimization can be substantially improved by unifying two complementary principles: (i) a projection-based subspace view that reduces gradient estimation variance by exploiting the intrinsic low-rank structure of model updates, and (ii) Muon-style spectral optimization that applies gradient orthogonalization to extract informative spectral structure from noisy ZO gradients. These findings form a unified framework of subspace gradient orthogonalization, which we instantiate in a new method, ZO-Muon, admitting a natural interpretation as a low-rank Muon optimizer in the ZO setting. Extensive experiments on large language models (LLMs) and vision transformers (ViTs) demonstrate that ZO-Muon significantly accelerates convergence and achieves a win-win improvement in accuracy and query/runtime efficiency. Notably, compared to the popular MeZO baseline, ZO-Muon requires only 24.7% of the queries to reach the same SST-2 performance for LLM fine-tuning, and improves accuracy by 25.1% on ViT-B fine-tuning on CIFAR-100.

Hanhui Wang, Yihua Zhang, Ruizheng Bai et al.

Recent advancements in diffusion models have made generative image editing more accessible, enabling creative edits but raising ethical concerns, particularly regarding malicious edits to human portraits that threaten privacy and identity security. Existing protection methods primarily rely on adversarial perturbations to nullify edits but often fail against diverse editing requests. We propose FaceLock, a novel approach to portrait protection that optimizes adversarial perturbations to destroy or significantly alter biometric information, rendering edited outputs biometrically unrecognizable. FaceLock integrates facial recognition and visual perception into perturbation optimization to provide robust protection against various editing attempts. We also highlight flaws in commonly used evaluation metrics and reveal how they can be manipulated, emphasizing the need for reliable assessments of protection. Experiments show FaceLock outperforms baselines in defending against malicious edits and is robust against purification techniques. Ablation studies confirm its stability and broad applicability across diffusion-based editing algorithms. Our work advances biometric defense and sets the foundation for privacy-preserving practices in image editing. The code is available at: https://github.com/taco-group/FaceLock.

Sijia Liu, Yicheng Lang, Soumyadeep Pal et al.

Zeroth-order (ZO) optimization, learning from finite differences of function evaluations without backpropagation, has recently regained attention in deep learning due to its memory efficiency and applicability to gray- or black-box pipelines. Yet, ZO methods are often dismissed as fundamentally unscalable because of estimator variance and unfavorable query complexity. We argue that this conclusion might be misguided: ZO optimization is underexplored, not underpowered. We show that many perceived limitations stem from myopic development practices, most notably full-space, element-wise, estimator-centric designs. We articulate six positions spanning the algorithmic, systems, and evaluation stack. First, we revisit the feasibility boundaries of estimator-centric ZO methods through variance control, variance-query tradeoffs, and directional-derivative lenses. Then, we identify three underexplored opportunities: (i) subspace and spectral views of ZO that enable interpretable variance reduction with graceful query scaling, (ii) the forward-only nature of ZO as a systems advantage for communication-efficient, pipeline-friendly, and resource-constrained training, and (iii) the need to de-obfuscate ZO evaluations from task complexity. We strongly advocate rethinking ZO optimization around its unique strengths and acting accordingly, opening a viable path toward large-scale, system-aware, and resource-efficient learning with ZO optimization.

Changchang Sun, Ren Wang, Yihua Zhang et al.

Machine unlearning (MU), which seeks to erase the influence of specific unwanted data from already-trained models, is becoming increasingly vital in model editing, particularly to comply with evolving data regulations like the ``right to be forgotten''. Conventional approaches are predominantly model-based, typically requiring retraining or fine-tuning the model's weights to meet unlearning requirements. In this work, we approach the MU problem from a novel input perturbation-based perspective, where the model weights remain intact throughout the unlearning process. We demonstrate the existence of a proactive input-based unlearning strategy, referred to forget vector, which can be generated as an input-agnostic data perturbation and remains as effective as model-based approximate unlearning approaches. We also explore forget vector arithmetic, whereby multiple class-specific forget vectors are combined through simple operations (e.g., linear combinations) to generate new forget vectors for unseen unlearning tasks, such as forgetting arbitrary subsets across classes. Extensive experiments validate the effectiveness and adaptability of the forget vector, showcasing its competitive performance relative to state-of-the-art model-based methods. Codes are available at https://github.com/Changchangsun/Forget-Vector.

Chongyu Fan, Yihua Zhang, Jinghan Jia et al.

Large reasoning models (LRMs), such as OpenAI's o1 and DeepSeek-R1, harness test-time scaling to perform multi-step reasoning for complex problem-solving. This reasoning process, executed before producing final answers, is often guided by special juncture tokens or textual segments that prompt self-evaluative reflection. We refer to these transition markers and reflective cues as "reflection tokens" (e.g., "wait", "but", "alternatively"). In this work, we treat reflection tokens as a "resource" and introduce the problem of resource allocation, aimed at improving the test-time compute performance of LRMs by adaptively regulating the frequency and placement of reflection tokens. Through empirical analysis, we show that both excessive and insufficient use of reflection tokens, referred to as over-reflection and under-reflection, can degrade model performance. To better understand and manage this trade-off, we draw an analogy between reflection token usage and learning rate scheduling in optimization. Building on this insight, we propose cyclical reflection token scheduling (termed CyclicReflex), a decoding strategy that dynamically modulates reflection token logits using a position-dependent triangular waveform. Experiments on MATH500, AIME2024/2025, and AMC2023 demonstrate that CyclicReflex consistently improves performance across model sizes (1.5B-8B), outperforming standard decoding and more recent approaches such as TIP (thought switching penalty) and S1. Codes are available at https://github.com/OPTML-Group/CyclicReflex.

Bingqi Shang, Yiwei Chen, Yihua Zhang et al.

Large language model (LLM) unlearning has become a critical mechanism for removing undesired data, knowledge, or behaviors from pre-trained models while retaining their general utility. Yet, with the rise of open-weight LLMs, we ask: can the unlearning process itself be backdoored, appearing successful under normal conditions yet reverting to pre-unlearned behavior when a hidden trigger is activated? Drawing inspiration from classical backdoor attacks that embed triggers into training data to enforce specific behaviors, we investigate backdoor unlearning, where models forget as intended in the clean setting but recover forgotten knowledge when the trigger appears. We show that designing such attacks presents unique challenges, hinging on where triggers are placed and how backdoor training is reinforced. We uncover a strong link between backdoor efficacy and the attention sink phenomenon, i.e., shallow input tokens consistently attract disproportionate attention in LLMs. Our analysis reveals that these attention sinks serve as gateways for backdoor unlearning: placing triggers at sink positions and aligning their attention values markedly enhances backdoor persistence. Extensive experiments validate these findings, showing that attention-sink-guided backdoor unlearning reliably restores forgotten knowledge in the presence of backdoor triggers, while behaving indistinguishably from a normally unlearned model when triggers are absent. Code is available at https://github.com/OPTML-Group/Unlearn-Backdoor.

Yihua Zhang, Guanhua Zhang, Prashant Khanduri et al.

Adversarial training (AT) is a widely recognized defense mechanism to gain the robustness of deep neural networks against adversarial attacks. It is built on min-max optimization (MMO), where the minimizer (i.e., defender) seeks a robust model to minimize the worst-case training loss in the presence of adversarial examples crafted by the maximizer (i.e., attacker). However, the conventional MMO method makes AT hard to scale. Thus, Fast-AT (Wong et al., 2020) and other recent algorithms attempt to simplify MMO by replacing its maximization step with the single gradient sign-based attack generation step. Although easy to implement, Fast-AT lacks theoretical guarantees, and its empirical performance is unsatisfactory due to the issue of robust catastrophic overfitting when training with strong adversaries. In this paper, we advance Fast-AT from the fresh perspective of bi-level optimization (BLO). We first show that the commonly-used Fast-AT is equivalent to using a stochastic gradient algorithm to solve a linearized BLO problem involving a sign operation. However, the discrete nature of the sign operation makes it difficult to understand the algorithm performance. Inspired by BLO, we design and analyze a new set of robust training algorithms termed Fast Bi-level AT (Fast-BAT), which effectively defends sign-based projected gradient descent (PGD) attacks without using any gradient sign method or explicit robust regularization. In practice, we show our method yields substantial robustness improvements over baselines across multiple models and datasets. Codes are available at https://github.com/OPTML-Group/Fast-BAT.

Renjie Gu, Jiazhen Du, Yihua Zhang et al.

Unlearning in large language models (LLMs) aims to remove harmful training data while preserving overall utility. However, we find that existing methods often hallucinate, generate abnormal token sequences, or behave inconsistently, raising safety and trust concerns. According to prior literature on LLM honesty, such behaviors are often associated with dishonesty. This motivates us to investigate the notion of honesty in the context of model unlearning. We propose a formal definition of unlearning honesty, which includes: (1) preserving both utility and honesty on retained knowledge, and (2) ensuring effective forgetting while encouraging the model to acknowledge its limitations and respond consistently to questions related to forgotten knowledge. To systematically evaluate the honesty of unlearning, we introduce a suite of metrics that cover utility, honesty on the retained set, effectiveness of forgetting, rejection rate and refusal stability in Q&A and MCQ settings. Evaluating 9 methods across 3 mainstream families shows that all current methods fail to meet these standards. After experimental and theoretical analyses, we present ReVa, a representation-alignment procedure that fine-tunes feature-randomized unlearned models to better acknowledge forgotten knowledge. On Q&A tasks from the forget set, ReVa achieves the highest rejection rate after two rounds of interaction, nearly doubling the performance of the second-best method. Remarkably, It also improves honesty on the retained set. We release our data and code at https://github.com/renjiegu.

Hongkang Li, Yihua Zhang, Shuai Zhang et al.

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).

Jinghan Jia, Jiancheng Liu, Yihua Zhang et al.

The need for effective unlearning mechanisms in large language models (LLMs) is increasingly urgent, driven by the necessity to adhere to data regulations and foster ethical generative AI practices. Despite growing interest of LLM unlearning, much of the existing research has focused on varied unlearning method designs to boost effectiveness and efficiency. However, the inherent relationship between model weights and LLM unlearning has not been extensively examined. In this paper, we systematically explore how model weights interact with unlearning processes in LLMs and we design the weight attribution-guided LLM unlearning method, WAGLE, which unveils the interconnections between 'influence' of weights and 'influence' of data to forget and retain in LLM generation. By strategically guiding the LLM unlearning across different types of unlearning methods and tasks, WAGLE can erase the undesired content, while maintaining the performance of the original tasks. We refer to the weight attribution-guided LLM unlearning method as WAGLE, which unveils the interconnections between 'influence' of weights and 'influence' of data to forget and retain in LLM generation. Our extensive experiments show that WAGLE boosts unlearning performance across a range of LLM unlearning methods such as gradient difference and (negative) preference optimization, applications such as fictitious unlearning, malicious use prevention, and copyrighted information removal, and models including Zephyr-7b-beta and Llama2-7b. To the best of our knowledge, our work offers the first principled method for attributing and pinpointing the influential weights in enhancing LLM unlearning. It stands in contrast to previous methods that lack weight attribution and simpler weight attribution techniques.

Yihua Zhang, Chongyu Fan, Yimeng Zhang et al.

The technological advancements in diffusion models (DMs) have demonstrated unprecedented capabilities in text-to-image generation and are widely used in diverse applications. However, they have also raised significant societal concerns, such as the generation of harmful content and copyright disputes. Machine unlearning (MU) has emerged as a promising solution, capable of removing undesired generative capabilities from DMs. However, existing MU evaluation systems present several key challenges that can result in incomplete and inaccurate assessments. To address these issues, we propose UnlearnCanvas, a comprehensive high-resolution stylized image dataset that facilitates the evaluation of the unlearning of artistic styles and associated objects. This dataset enables the establishment of a standardized, automated evaluation framework with 7 quantitative metrics assessing various aspects of the unlearning performance for DMs. Through extensive experiments, we benchmark 9 state-of-the-art MU methods for DMs, revealing novel insights into their strengths, weaknesses, and underlying mechanisms. Additionally, we explore challenging unlearning scenarios for DMs to evaluate worst-case performance against adversarial prompts, the unlearning of finer-scale concepts, and sequential unlearning. We hope that this study can pave the way for developing more effective, accurate, and robust DM unlearning methods, ensuring safer and more ethical applications of DMs in the future. The dataset, benchmark, and codes are publicly available at https://unlearn-canvas.netlify.app/.

Hongkang Li, Shuai Zhang, Yihua Zhang et al.

Group imbalance has been a known problem in empirical risk minimization (ERM), where the achieved high average accuracy is accompanied by low accuracy in a minority group. Despite algorithmic efforts to improve the minority group accuracy, a theoretical generalization analysis of ERM on individual groups remains elusive. By formulating the group imbalance problem with the Gaussian Mixture Model, this paper quantifies the impact of individual groups on the sample complexity, the convergence rate, and the average and group-level testing performance. Although our theoretical framework is centered on binary classification using a one-hidden-layer neural network, to the best of our knowledge, we provide the first theoretical analysis of the group-level generalization of ERM in addition to the commonly studied average generalization performance. Sample insights of our theoretical results include that when all group-level co-variance is in the medium regime and all mean are close to zero, the learning performance is most desirable in the sense of a small sample complexity, a fast training rate, and a high average and group-level testing accuracy. Moreover, we show that increasing the fraction of the minority group in the training data does not necessarily improve the generalization performance of the minority group. Our theoretical results are validated on both synthetic and empirical datasets, such as CelebA and CIFAR-10 in image classification.

Changsheng Wang, Chongyu Fan, Yihua Zhang et al.

Recent advances in large reasoning models (LRMs) have enabled strong chain-of-thought (CoT) generation through test-time computation. While these multi-step reasoning capabilities represent a major milestone in language model performance, they also introduce new safety risks. In this work, we present the first systematic study to revisit the problem of machine unlearning in the context of LRMs. Machine unlearning refers to the process of removing the influence of sensitive, harmful, or undesired data or knowledge from a trained model without full retraining. We show that conventional unlearning algorithms, originally designed for non-reasoning models, are inadequate for LRMs. In particular, even when final answers are successfully erased, sensitive information often persists within the intermediate reasoning steps, i.e., CoT trajectories. To address this challenge, we extend conventional unlearning and propose Reasoning-aware Representation Misdirection for Unlearning ($R^2MU$), a novel method that effectively suppresses sensitive reasoning traces and prevents the generation of associated final answers, while preserving the model's reasoning ability. Our experiments demonstrate that $R^2MU$ significantly reduces sensitive information leakage within reasoning traces and achieves strong performance across both safety and reasoning benchmarks, evaluated on state-of-the-art models such as DeepSeek-R1-Distill-LLaMA-8B and DeepSeek-R1-Distill-Qwen-14B.

Haomin Zhuang, Yihua Zhang, Kehan Guo et al.

Recent advancements in LLMs unlearning have shown remarkable success in removing unwanted data-model influences while preserving the model's utility for legitimate knowledge. Despite these strides, sparse Mixture-of-Experts (MoE) LLMs--a key subset of the LLM family--have remained unexplored in the context of unlearning. As MoE LLMs are celebrated for their exceptional performance, we ask:How can unlearning be performed effectively and efficiently on MoE LLMs? Our pilot study shows that the dynamic routing nature of MoE LLMs introduces unique challenges, leading to excessive forgetting, uncontrolled knowledge erasure and substantial utility drops when existing unlearning methods are applied. To address this, we propose a novel Selected-Expert Unlearning Framework (SEUF). Through expert attribution, unlearning is concentrated on the most actively engaged experts for the specified knowledge. Concurrently, an anchor loss is applied to the router to stabilize the active state of this targeted expert, ensuring focused and controlled unlearning. SEUF is compatible with various standard unlearning algorithms. Extensive experiments demonstrate that SEUF enhances both forget quality up to 5% and model utility by 35% on MoE LLMs across various benchmarks and LLM architectures (compared to standard unlearning algorithms), while only unlearning 0.06% of the model parameters.

Yiwei Chen, Yuguang Yao, Yihua Zhang et al.

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 these issues, 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 MU-based alignment reduces the attack success rate by up to 60.17% and cuts unnecessary rejections by over 84.20%. WARNING: There exist AI generations that may be offensive in nature.

Changsheng Wang, Yihua Zhang, Jinghan Jia et al.

Machine unlearning offers a promising solution to privacy and safety concerns in large language models (LLMs) by selectively removing targeted knowledge while preserving utility. However, current methods are highly sensitive to downstream fine-tuning, which can quickly recover forgotten information-even from unrelated tasks. To address this, we introduce invariance into unlearning for the first time, inspired by invariant risk minimization (IRM). Building on this principle, we propose invariant LLM unlearning (ILU), a regularization-based framework that enhances robustness. Notably, ILU generalizes well to diverse fine-tuning tasks, even when trained using a single dataset. A task vector analysis is also provided to further elucidate the rationale behind ILU's effectiveness. Extensive experiments on the WMDP and MUSE benchmark, reveal that ILU significantly outperforms state-of-the-art unlearning methods, including negative preference optimization (NPO) and representation misdirection for unlearning (RMU). Notably, ILU achieves superior unlearning robustness across diverse downstream fine-tuning scenarios (e.g., math, paraphrase detection, and sentiment analysis) while preserving the fine-tuning performance.

Mohammad Jafari, Yimeng Zhang, Yihua Zhang et al.

As machine learning tasks continue to evolve, the trend has been to gather larger datasets and train increasingly larger models. While this has led to advancements in accuracy, it has also escalated computational costs to unsustainable levels. Addressing this, our work aims to strike a delicate balance between computational efficiency and model accuracy, a persisting challenge in the field. We introduce a novel method that employs core subset selection for reweighting, effectively optimizing both computational time and model performance. By focusing on a strategically selected coreset, our approach offers a robust representation, as it efficiently minimizes the influence of outliers. The re-calibrated weights are then mapped back to and propagated across the entire dataset. Our experimental results substantiate the effectiveness of this approach, underscoring its potential as a scalable and precise solution for model training.

Mohan Zhang, Yihua Zhang, Jinghan Jia et al.

Modern large reasoning models (LRMs) exhibit impressive multi-step problem-solving via chain-of-thought (CoT) reasoning. However, this iterative thinking mechanism introduces a new vulnerability surface. We present the Deadlock Attack, a resource exhaustion method that hijacks an LRM's generative control flow by training a malicious adversarial embedding to induce perpetual reasoning loops. Specifically, the optimized embedding encourages transitional tokens (e.g., "Wait", "But") after reasoning steps, preventing the model from concluding its answer. A key challenge we identify is the continuous-to-discrete projection gap: naïve projections of adversarial embeddings to token sequences nullify the attack. To overcome this, we introduce a backdoor implantation strategy, enabling reliable activation through specific trigger tokens. Our method achieves a 100% attack success rate across four advanced LRMs (Phi-RM, Nemotron-Nano, R1-Qwen, R1-Llama) and three math reasoning benchmarks, forcing models to generate up to their maximum token limits. The attack is also stealthy (in terms of causing negligible utility loss on benign user inputs) and remains robust against existing strategies trying to mitigate the overthinking issue. Our findings expose a critical and underexplored security vulnerability in LRMs from the perspective of reasoning (in)efficiency.

Changsheng Wang, Yihua Zhang, Dennis Wei et al.

Large language models (LLMs) exhibit remarkable generative capabilities but raise ethical and security concerns by memorizing sensitive data, reinforcing biases, and producing harmful content. These risks have spurred interest in LLM unlearning, the task of removing knowledge associated with undesirable data from pre-trained models. However, most existing methods assume access to clean, well-defined forget data samples, whereas real-world forget data could often be low-quality, synthetically rewritten, or watermarked, casting doubt on the reliability of unlearning. This work presents the first study of unlearning under perturbed or low-fidelity forget data, referred to as noisy forget sets. By systematically benchmarking state-of-the-art LLM unlearning methods, RMU and NPO, on such noisy forget sets, we find that unlearning remains surprisingly robust to perturbations, provided that core semantic signals are preserved. To explain this robustness, we propose a saliency-based interpretation: key semantic components that drive forgetting remain consistently influential despite substantial variation in surface form. This suggests that unlearning algorithms are primarily guided by deep semantic cues rather than shallow lexical patterns.

Yicheng Lang, Yihua Zhang, Chongyu Fan et al.

Large language model (LLM) unlearning aims to surgically remove the influence of undesired data or knowledge from an existing model while preserving its utility on unrelated tasks. This paradigm has shown promise in addressing privacy and safety concerns. However, recent findings reveal that unlearning effects are often fragile: post-unlearning manipulations such as weight quantization or fine-tuning can quickly neutralize the intended forgetting. Prior efforts to improve robustness primarily reformulate unlearning objectives by explicitly assuming the role of vulnerability sources. In this work, we take a different perspective by investigating the role of the optimizer, independent of unlearning objectives and formulations, in shaping unlearning robustness. We show that the 'grade' of the optimizer, defined by the level of information it exploits, ranging from zeroth-order (gradient-free) to first-order (gradient-based) to second-order (Hessian-based), is tightly linked to the resilience of unlearning. Surprisingly, we find that downgrading the optimizer, such as using zeroth-order methods or compressed-gradient variants (e.g., gradient sign-based optimizers), often leads to stronger robustness. While these optimizers produce noisier and less precise updates, they encourage convergence to harder-to-disturb basins in the loss landscape, thereby resisting post-training perturbations. By connecting zeroth-order methods with randomized smoothing, we further highlight their natural advantage for robust unlearning. Motivated by these insights, we propose a hybrid optimizer that combines first-order and zeroth-order updates, preserving unlearning efficacy while enhancing robustness. Extensive experiments on the MUSE and WMDP benchmarks, across multiple LLM unlearning algorithms, validate that our approach achieves more resilient forgetting without sacrificing unlearning quality.

Yuhao Sun, Yihua Zhang, Gaowen Liu et al.

With the increasing demand for the right to be forgotten, machine unlearning (MU) has emerged as a vital tool for enhancing trust and regulatory compliance by enabling the removal of sensitive data influences from machine learning (ML) models. However, most MU algorithms primarily rely on in-training methods to adjust model weights, with limited exploration of the benefits that data-level adjustments could bring to the unlearning process. To address this gap, we propose a novel approach that leverages digital watermarking to facilitate MU by strategically modifying data content. By integrating watermarking, we establish a controlled unlearning mechanism that enables precise removal of specified data while maintaining model utility for unrelated tasks. We first examine the impact of watermarked data on MU, finding that MU effectively generalizes to watermarked data. Building on this, we introduce an unlearning-friendly watermarking framework, termed Water4MU, to enhance unlearning effectiveness. The core of Water4MU is a bi-level optimization (BLO) framework: at the upper level, the watermarking network is optimized to minimize unlearning difficulty, while at the lower level, the model itself is trained independently of watermarking. Experimental results demonstrate that Water4MU is effective in MU across both image classification and image generation tasks. Notably, it outperforms existing methods in challenging MU scenarios, known as "challenging forgets".

Yihua Zhang, Marina Blanton, Ghada Almashaqbeh

Recent compilers allow a general-purpose program (written in a conventional programming language) that handles private data to be translated into secure distributed implementation of the corresponding functionality. The resulting program is then guaranteed to provably protect private data using secure multi-party computation techniques. The goals of such compilers are generality, usability, and efficiency, but the complete set of features of a modern programming language has not been supported to date by the existing compilers. In particular, recent compilers PICCO and the two-party ANSI C compiler strive to translate any C program into its secure multi-party implementation, but currently lack support for pointers and dynamic memory allocation, which are important components of many C programs. In this work, we mitigate the limitation and add support for pointers to private data and consequently dynamic memory allocation to the PICCO compiler, enabling it to handle a more diverse set of programs over private data. Because doing so opens up a new design space, we investigate the use of pointers to private data (with known as well as private locations stored in them) in programs and report our findings. Besides dynamic memory allocation, we examine other important topics associated with common pointer use such as reference by pointer/address, casting, and building various data structures in the context of secure multi-party computation. This results in enabling the compiler to automatically translate a user program that uses pointers to private data into its distributed implementation that provably protects private data throughout the computation. We empirically evaluate the constructions and report on performance of representative programs.