Abstract:In this study, we investigate the vulnerability of image watermarks to diffusion-model-based image editing, a challenge exacerbated by the computational cost of accessing gradient information and the closed-source nature of many diffusion models. To address this issue, we introduce JIGMARK. This first-of-its-kind watermarking technique enhances robustness through contrastive learning with pairs of images, processed and unprocessed by diffusion models, without needing a direct backpropagation of the diffusion process. Our evaluation reveals that JIGMARK significantly surpasses existing watermarking solutions in resilience to diffusion-model edits, demonstrating a True Positive Rate more than triple that of leading baselines at a 1% False Positive Rate while preserving image quality. At the same time, it consistently improves the robustness against other conventional perturbations (like JPEG, blurring, etc.) and malicious watermark attacks over the state-of-the-art, often by a large margin. Furthermore, we propose the Human Aligned Variation (HAV) score, a new metric that surpasses traditional similarity measures in quantifying the number of image derivatives from image editing.
Abstract:As a defense strategy against adversarial attacks, adversarial detection aims to identify and filter out adversarial data from the data flow based on discrepancies in distribution and noise patterns between natural and adversarial data. Although previous detection methods achieve high performance in detecting gradient-based adversarial attacks, new attacks based on generative models with imbalanced and anisotropic noise patterns evade detection. Even worse, existing techniques either necessitate access to attack data before deploying a defense or incur a significant time cost for inference, rendering them impractical for defending against newly emerging attacks that are unseen by defenders. In this paper, we explore the proximity relationship between adversarial noise distributions and demonstrate the existence of an open covering for them. By learning to distinguish this open covering from the distribution of natural data, we can develop a detector with strong generalization capabilities against all types of adversarial attacks. Based on this insight, we heuristically propose Perturbation Forgery, which includes noise distribution perturbation, sparse mask generation, and pseudo-adversarial data production, to train an adversarial detector capable of detecting unseen gradient-based, generative-model-based, and physical adversarial attacks, while remaining agnostic to any specific models. Comprehensive experiments conducted on multiple general and facial datasets, with a wide spectrum of attacks, validate the strong generalization of our method.
Abstract:Fact tracing seeks to identify specific training examples that serve as the knowledge source for a given query. Existing approaches to fact tracing rely on assessing the similarity between each training sample and the query along a certain dimension, such as lexical similarity, gradient, or embedding space. However, these methods fall short of effectively distinguishing between samples that are merely relevant and those that actually provide supportive evidence for the information sought by the query. This limitation often results in suboptimal effectiveness. Moreover, these approaches necessitate the examination of the similarity of individual training points for each query, imposing significant computational demands and creating a substantial barrier for practical applications. This paper introduces FASTTRACK, a novel approach that harnesses the capabilities of Large Language Models (LLMs) to validate supportive evidence for queries and at the same time clusters the training database towards a reduced extent for LLMs to trace facts. Our experiments show that FASTTRACK substantially outperforms existing methods in both accuracy and efficiency, achieving more than 100\% improvement in F1 score over the state-of-the-art methods while being X33 faster than \texttt{TracIn}.
Abstract:While recent progress in video-text retrieval has been driven by the exploration of powerful model architectures and training strategies, the representation learning ability of video-text retrieval models is still limited due to low-quality and scarce training data annotations. To address this issue, we present a novel video-text learning paradigm, HaVTR, which augments video and text data to learn more generalized features. Specifically, we first adopt a simple augmentation method, which generates self-similar data by randomly duplicating or dropping subwords and frames. In addition, inspired by the recent advancement in visual and language generative models, we propose a more powerful augmentation method through textual paraphrasing and video stylization using large language models (LLMs) and visual generative models (VGMs). Further, to bring richer information into video and text, we propose a hallucination-based augmentation method, where we use LLMs and VGMs to generate and add new relevant information to the original data. Benefiting from the enriched data, extensive experiments on several video-text retrieval benchmarks demonstrate the superiority of HaVTR over existing methods.
Abstract:Generative Adversarial Networks (GANs) have emerged as powerful tools not only for high-quality image generation but also for real image editing through manipulation of their interpretable latent spaces. Recent advancements in GANs include the development of 3D-aware models such as EG3D, characterized by efficient triplane-based architectures enabling the reconstruction of 3D geometry from single images. However, scant attention has been devoted to providing an integrated framework for high-quality reference-based 3D-aware image editing within this domain. This study addresses this gap by exploring and demonstrating the effectiveness of EG3D's triplane space for achieving advanced reference-based edits, presenting a unique perspective on 3D-aware image editing through our novel pipeline. Our approach integrates the encoding of triplane features, spatial disentanglement and automatic localization of features in the triplane domain, and fusion learning for desired image editing. Moreover, our framework demonstrates versatility across domains, extending its effectiveness to animal face edits and partial stylization of cartoon portraits. The method shows significant improvements over relevant 3D-aware latent editing and 2D reference-based editing methods, both qualitatively and quantitatively. Project page: https://three-bee.github.io/triplane_edit
Abstract:Recent advancements in Large Language Models (LLMs) have showcased remarkable capabilities across various tasks in different domains. However, the emergence of biases and the potential for generating harmful content in LLMs, particularly under malicious inputs, pose significant challenges. Current mitigation strategies, while effective, are not resilient under adversarial attacks. This paper introduces Resilient Guardrails for Large Language Models (RigorLLM), a novel framework designed to efficiently and effectively moderate harmful and unsafe inputs and outputs for LLMs. By employing a multi-faceted approach that includes energy-based training data augmentation through Langevin dynamics, optimizing a safe suffix for inputs via minimax optimization, and integrating a fusion-based model combining robust KNN with LLMs based on our data augmentation, RigorLLM offers a robust solution to harmful content moderation. Our experimental evaluations demonstrate that RigorLLM not only outperforms existing baselines like OpenAI API and Perspective API in detecting harmful content but also exhibits unparalleled resilience to jailbreaking attacks. The innovative use of constrained optimization and a fusion-based guardrail approach represents a significant step forward in developing more secure and reliable LLMs, setting a new standard for content moderation frameworks in the face of evolving digital threats.
Abstract:Despite the impressive capabilities of large language models (LLMs) across diverse applications, they still suffer from trustworthiness issues, such as hallucinations and misalignments. Retrieval-augmented language models (RAG) have been proposed to enhance the credibility of generations by grounding external knowledge, but the theoretical understandings of their generation risks remains unexplored. In this paper, we answer: 1) whether RAG can indeed lead to low generation risks, 2) how to provide provable guarantees on the generation risks of RAG and vanilla LLMs, and 3) what sufficient conditions enable RAG models to reduce generation risks. We propose C-RAG, the first framework to certify generation risks for RAG models. Specifically, we provide conformal risk analysis for RAG models and certify an upper confidence bound of generation risks, which we refer to as conformal generation risk. We also provide theoretical guarantees on conformal generation risks for general bounded risk functions under test distribution shifts. We prove that RAG achieves a lower conformal generation risk than that of a single LLM when the quality of the retrieval model and transformer is non-trivial. Our intensive empirical results demonstrate the soundness and tightness of our conformal generation risk guarantees across four widely-used NLP datasets on four state-of-the-art retrieval models.
Abstract:Natural language serves as a common and straightforward control signal for humans to interact seamlessly with machines. Recognizing the importance of this interface, the machine learning community is investing considerable effort in generating data that is semantically coherent with textual instructions. While strides have been made in text-to-data generation spanning image editing, audio synthesis, video creation, and beyond, low-resource areas characterized by expensive annotations or complex data structures, such as molecules, motion dynamics, and time series, often lack textual labels. This deficiency impedes supervised learning, thereby constraining the application of advanced generative models for text-to-data tasks. In response to these challenges in the low-resource scenario, we propose Text2Data, a novel approach that utilizes unlabeled data to understand the underlying data distribution through an unsupervised diffusion model. Subsequently, it undergoes controllable finetuning via a novel constraint optimization-based learning objective that ensures controllability and effectively counteracts catastrophic forgetting. Comprehensive experiments demonstrate that Text2Data is able to achieve enhanced performance regarding controllability across various modalities, including molecules, motions and time series, when compared to existing baselines.
Abstract:Vision-Language Models (VLMs) excel in generating textual responses from visual inputs, yet their versatility raises significant security concerns. This study takes the first step in exposing VLMs' susceptibility to data poisoning attacks that can manipulate responses to innocuous, everyday prompts. We introduce Shadowcast, a stealthy data poisoning attack method where poison samples are visually indistinguishable from benign images with matching texts. Shadowcast demonstrates effectiveness in two attack types. The first is Label Attack, tricking VLMs into misidentifying class labels, such as confusing Donald Trump for Joe Biden. The second is Persuasion Attack, which leverages VLMs' text generation capabilities to craft narratives, such as portraying junk food as health food, through persuasive and seemingly rational descriptions. We show that Shadowcast are highly effective in achieving attacker's intentions using as few as 50 poison samples. Moreover, these poison samples remain effective across various prompts and are transferable across different VLM architectures in the black-box setting. This work reveals how poisoned VLMs can generate convincing yet deceptive misinformation and underscores the importance of data quality for responsible deployments of VLMs. Our code is available at: https://github.com/umd-huang-lab/VLM-Poisoning.
Abstract:In response to the rapidly evolving nature of adversarial attacks on a monthly basis, numerous defenses have been proposed to generalize against as many known attacks as possible. However, designing a defense method that can generalize to all types of attacks, including unseen ones, is not realistic because the environment in which defense systems operate is dynamic and comprises various unique attacks used by many attackers. The defense system needs to upgrade itself by utilizing few-shot defense feedback and efficient memory. Therefore, we propose the first continual adversarial defense (CAD) framework that adapts to any attacks in a dynamic scenario, where various attacks emerge stage by stage. In practice, CAD is modeled under four principles: (1) continual adaptation to new attacks without catastrophic forgetting, (2) few-shot adaptation, (3) memory-efficient adaptation, and (4) high accuracy on both clean and adversarial images. We leverage cutting-edge continual learning, few-shot learning, and ensemble learning techniques to qualify the principles. Experiments conducted on CIFAR-10 and ImageNet-100 validate the effectiveness of our approach against multiple stages of 10 modern adversarial attacks and significant improvements over 10 baseline methods. In particular, CAD is capable of quickly adapting with minimal feedback and a low cost of defense failure, while maintaining good performance against old attacks. Our research sheds light on a brand-new paradigm for continual defense adaptation against dynamic and evolving attacks.