Unveiling Concept Attribution in Diffusion Models

Diffusion models have shown remarkable abilities in generating realistic and high-quality images from text prompts. However, a trained model remains black-box; little do we know about the role of its components in exhibiting a concept such as objects or styles. Recent works employ causal tracing to localize layers storing knowledge in generative models without showing how those layers contribute to the target concept. In this work, we approach the model interpretability problem from a more general perspective and pose a question: \textit{``How do model components work jointly to demonstrate knowledge?''}. We adapt component attribution to decompose diffusion models, unveiling how a component contributes to a concept. Our framework allows effective model editing, in particular, we can erase a concept from diffusion models by removing positive components while remaining knowledge of other concepts. Surprisingly, we also show there exist components that contribute negatively to a concept, which has not been discovered in the knowledge localization approach. Experimental results confirm the role of positive and negative components pinpointed by our framework, depicting a complete view of interpreting generative models. Our code is available at \url{https://github.com/mail-research/CAD-attribution4diffusion}

Flatness-aware Sequential Learning Generates Resilient Backdoors

Recently, backdoor attacks have become an emerging threat to the security of machine learning models. From the adversary's perspective, the implanted backdoors should be resistant to defensive algorithms, but some recently proposed fine-tuning defenses can remove these backdoors with notable efficacy. This is mainly due to the catastrophic forgetting (CF) property of deep neural networks. This paper counters CF of backdoors by leveraging continual learning (CL) techniques. We begin by investigating the connectivity between a backdoored and fine-tuned model in the loss landscape. Our analysis confirms that fine-tuning defenses, especially the more advanced ones, can easily push a poisoned model out of the backdoor regions, making it forget all about the backdoors. Based on this finding, we re-formulate backdoor training through the lens of CL and propose a novel framework, named Sequential Backdoor Learning (SBL), that can generate resilient backdoors. This framework separates the backdoor poisoning process into two tasks: the first task learns a backdoored model, while the second task, based on the CL principles, moves it to a backdoored region resistant to fine-tuning. We additionally propose to seek flatter backdoor regions via a sharpness-aware minimizer in the framework, further strengthening the durability of the implanted backdoor. Finally, we demonstrate the effectiveness of our method through extensive empirical experiments on several benchmark datasets in the backdoor domain. The source code is available at https://github.com/mail-research/SBL-resilient-backdoors

Less is More: Sparse Watermarking in LLMs with Enhanced Text Quality

With the widespread adoption of Large Language Models (LLMs), concerns about potential misuse have emerged. To this end, watermarking has been adapted to LLM, enabling a simple and effective way to detect and monitor generated text. However, while the existing methods can differentiate between watermarked and unwatermarked text with high accuracy, they often face a trade-off between the quality of the generated text and the effectiveness of the watermarking process. In this work, we present a novel type of LLM watermark, Sparse Watermark, which aims to mitigate this trade-off by applying watermarks to a small subset of generated tokens distributed across the text. The key strategy involves anchoring watermarked tokens to words that have specific Part-of-Speech (POS) tags. Our experimental results demonstrate that the proposed watermarking scheme achieves high detectability while generating text that outperforms previous LLM watermarking methods in quality across various tasks

Wicked Oddities: Selectively Poisoning for Effective Clean-Label Backdoor Attacks

Deep neural networks are vulnerable to backdoor attacks, a type of adversarial attack that poisons the training data to manipulate the behavior of models trained on such data. Clean-label attacks are a more stealthy form of backdoor attacks that can perform the attack without changing the labels of poisoned data. Early works on clean-label attacks added triggers to a random subset of the training set, ignoring the fact that samples contribute unequally to the attack's success. This results in high poisoning rates and low attack success rates. To alleviate the problem, several supervised learning-based sample selection strategies have been proposed. However, these methods assume access to the entire labeled training set and require training, which is expensive and may not always be practical. This work studies a new and more practical (but also more challenging) threat model where the attacker only provides data for the target class (e.g., in face recognition systems) and has no knowledge of the victim model or any other classes in the training set. We study different strategies for selectively poisoning a small set of training samples in the target class to boost the attack success rate in this setting. Our threat model poses a serious threat in training machine learning models with third-party datasets, since the attack can be performed effectively with limited information. Experiments on benchmark datasets illustrate the effectiveness of our strategies in improving clean-label backdoor attacks.

MetaLLM: A High-performant and Cost-efficient Dynamic Framework for Wrapping LLMs

The rapid progress in machine learning (ML) has brought forth many large language models (LLMs) that excel in various tasks and areas. These LLMs come with different abilities and costs in terms of computation or pricing. Since the demand for each query can vary, e.g., because of the queried domain or its complexity, defaulting to one LLM in an application is not usually the best choice, whether it is the biggest, priciest, or even the one with the best average test performance. Consequently, picking the right LLM that is both accurate and cost-effective for an application remains a challenge. In this paper, we introduce MetaLLM, a framework that dynamically and intelligently routes each query to the optimal LLM (among several available LLMs) for classification tasks, achieving significantly improved accuracy and cost-effectiveness. By framing the selection problem as a multi-armed bandit, MetaLLM balances prediction accuracy and cost efficiency under uncertainty. Our experiments, conducted on popular LLM platforms such as OpenAI's GPT models, Amazon's Titan, Anthropic's Claude, and Meta's LLaMa, showcase MetaLLM's efficacy in real-world scenarios, laying the groundwork for future extensions beyond classification tasks.

Forget but Recall: Incremental Latent Rectification in Continual Learning

Intrinsic capability to continuously learn a changing data stream is a desideratum of deep neural networks (DNNs). However, current DNNs suffer from catastrophic forgetting, which hinders remembering past knowledge. To mitigate this issue, existing Continual Learning (CL) approaches either retain exemplars for replay, regularize learning, or allocate dedicated capacity for new tasks. This paper investigates an unexplored CL direction for incremental learning called Incremental Latent Rectification or ILR. In a nutshell, ILR learns to propagate with correction (or rectify) the representation from the current trained DNN backward to the representation space of the old task, where performing predictive decisions is easier. This rectification process only employs a chain of small representation mapping networks, called rectifier units. Empirical experiments on several continual learning benchmarks, including CIFAR10, CIFAR100, and Tiny ImageNet, demonstrate the effectiveness and potential of this novel CL direction compared to existing representative CL methods.

Fooling the Textual Fooler via Randomizing Latent Representations

Despite outstanding performance in a variety of NLP tasks, recent studies have revealed that NLP models are vulnerable to adversarial attacks that slightly perturb the input to cause the models to misbehave. Among these attacks, adversarial word-level perturbations are well-studied and effective attack strategies. Since these attacks work in black-box settings, they do not require access to the model architecture or model parameters and thus can be detrimental to existing NLP applications. To perform an attack, the adversary queries the victim model many times to determine the most important words in an input text and to replace these words with their corresponding synonyms. In this work, we propose a lightweight and attack-agnostic defense whose main goal is to perplex the process of generating an adversarial example in these query-based black-box attacks; that is to fool the textual fooler. This defense, named AdvFooler, works by randomizing the latent representation of the input at inference time. Different from existing defenses, AdvFooler does not necessitate additional computational overhead during training nor relies on assumptions about the potential adversarial perturbation set while having a negligible impact on the model's accuracy. Our theoretical and empirical analyses highlight the significance of robustness resulting from confusing the adversary via randomizing the latent space, as well as the impact of randomization on clean accuracy. Finally, we empirically demonstrate near state-of-the-art robustness of AdvFooler against representative adversarial word-level attacks on two benchmark datasets.

Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks

Recent works have shown that deep neural networks are vulnerable to adversarial examples that find samples close to the original image but can make the model misclassify. Even with access only to the model's output, an attacker can employ black-box attacks to generate such adversarial examples. In this work, we propose a simple and lightweight defense against black-box attacks by adding random noise to hidden features at intermediate layers of the model at inference time. Our theoretical analysis confirms that this method effectively enhances the model's resilience against both score-based and decision-based black-box attacks. Importantly, our defense does not necessitate adversarial training and has minimal impact on accuracy, rendering it applicable to any pre-trained model. Our analysis also reveals the significance of selectively adding noise to different parts of the model based on the gradient of the adversarial objective function, which can be varied during the attack. We demonstrate the robustness of our defense against multiple black-box attacks through extensive empirical experiments involving diverse models with various architectures.

Everyone Can Attack: Repurpose Lossy Compression as a Natural Backdoor Attack

The vulnerabilities to backdoor attacks have recently threatened the trustworthiness of machine learning models in practical applications. Conventional wisdom suggests that not everyone can be an attacker since the process of designing the trigger generation algorithm often involves significant effort and extensive experimentation to ensure the attack's stealthiness and effectiveness. Alternatively, this paper shows that there exists a more severe backdoor threat: anyone can exploit an easily-accessible algorithm for silent backdoor attacks. Specifically, this attacker can employ the widely-used lossy image compression from a plethora of compression tools to effortlessly inject a trigger pattern into an image without leaving any noticeable trace; i.e., the generated triggers are natural artifacts. One does not require extensive knowledge to click on the "convert" or "save as" button while using tools for lossy image compression. Via this attack, the adversary does not need to design a trigger generator as seen in prior works and only requires poisoning the data. Empirically, the proposed attack consistently achieves 100% attack success rate in several benchmark datasets such as MNIST, CIFAR-10, GTSRB and CelebA. More significantly, the proposed attack can still achieve almost 100% attack success rate with very small (approximately 10%) poisoning rates in the clean label setting. The generated trigger of the proposed attack using one lossy compression algorithm is also transferable across other related compression algorithms, exacerbating the severity of this backdoor threat. This work takes another crucial step toward understanding the extensive risks of backdoor attacks in practice, urging practitioners to investigate similar attacks and relevant backdoor mitigation methods.

Marksman Backdoor: Backdoor Attacks with Arbitrary Target Class

In recent years, machine learning models have been shown to be vulnerable to backdoor attacks. Under such attacks, an adversary embeds a stealthy backdoor into the trained model such that the compromised models will behave normally on clean inputs but will misclassify according to the adversary's control on maliciously constructed input with a trigger. While these existing attacks are very effective, the adversary's capability is limited: given an input, these attacks can only cause the model to misclassify toward a single pre-defined or target class. In contrast, this paper exploits a novel backdoor attack with a much more powerful payload, denoted as Marksman, where the adversary can arbitrarily choose which target class the model will misclassify given any input during inference. To achieve this goal, we propose to represent the trigger function as a class-conditional generative model and to inject the backdoor in a constrained optimization framework, where the trigger function learns to generate an optimal trigger pattern to attack any target class at will while simultaneously embedding this generative backdoor into the trained model. Given the learned trigger-generation function, during inference, the adversary can specify an arbitrary backdoor attack target class, and an appropriate trigger causing the model to classify toward this target class is created accordingly. We show empirically that the proposed framework achieves high attack performance while preserving the clean-data performance in several benchmark datasets, including MNIST, CIFAR10, GTSRB, and TinyImageNet. The proposed Marksman backdoor attack can also easily bypass existing backdoor defenses that were originally designed against backdoor attacks with a single target class. Our work takes another significant step toward understanding the extensive risks of backdoor attacks in practice.