Bo Li

Boxin Wang, Weixin Chen, Hengzhi Pei et al. · berkeley, microsoft-research

Generative Pre-trained Transformer (GPT) models have exhibited exciting progress in their capabilities, capturing the interest of practitioners and the public alike. Yet, while the literature on the trustworthiness of GPT models remains limited, practitioners have proposed employing capable GPT models for sensitive applications such as healthcare and finance -- where mistakes can be costly. To this end, this work proposes a comprehensive trustworthiness evaluation for large language models with a focus on GPT-4 and GPT-3.5, considering diverse perspectives -- including toxicity, stereotype bias, adversarial robustness, out-of-distribution robustness, robustness on adversarial demonstrations, privacy, machine ethics, and fairness. Based on our evaluations, we discover previously unpublished vulnerabilities to trustworthiness threats. For instance, we find that GPT models can be easily misled to generate toxic and biased outputs and leak private information in both training data and conversation history. We also find that although GPT-4 is usually more trustworthy than GPT-3.5 on standard benchmarks, GPT-4 is more vulnerable given jailbreaking system or user prompts, potentially because GPT-4 follows (misleading) instructions more precisely. Our work illustrates a comprehensive trustworthiness evaluation of GPT models and sheds light on the trustworthiness gaps. Our benchmark is publicly available at https://decodingtrust.github.io/ ; our dataset can be previewed at https://huggingface.co/datasets/AI-Secure/DecodingTrust ; a concise version of this work is at https://openreview.net/pdf?id=kaHpo8OZw2 .

Huan Zhang, Shiqi Wang, Kaidi Xu et al. · amazon-science

Bound propagation methods, when combined with branch and bound, are among the most effective methods to formally verify properties of deep neural networks such as correctness, robustness, and safety. However, existing works cannot handle the general form of cutting plane constraints widely accepted in traditional solvers, which are crucial for strengthening verifiers with tightened convex relaxations. In this paper, we generalize the bound propagation procedure to allow the addition of arbitrary cutting plane constraints, including those involving relaxed integer variables that do not appear in existing bound propagation formulations. Our generalized bound propagation method, GCP-CROWN, opens up the opportunity to apply general cutting plane methods for neural network verification while benefiting from the efficiency and GPU acceleration of bound propagation methods. As a case study, we investigate the use of cutting planes generated by off-the-shelf mixed integer programming (MIP) solver. We find that MIP solvers can generate high-quality cutting planes for strengthening bound-propagation-based verifiers using our new formulation. Since the branching-focused bound propagation procedure and the cutting-plane-focused MIP solver can run in parallel utilizing different types of hardware (GPUs and CPUs), their combination can quickly explore a large number of branches with strong cutting planes, leading to strong verification performance. Experiments demonstrate that our method is the first verifier that can completely solve the oval20 benchmark and verify twice as many instances on the oval21 benchmark compared to the best tool in VNN-COMP 2021, and also noticeably outperforms state-of-the-art verifiers on a wide range of benchmarks. GCP-CROWN is part of the $α,\!β$-CROWN verifier, the VNN-COMP 2022 winner. Code is available at http://PaperCode.cc/GCP-CROWN

Yiming Li, Yang Bai, Yong Jiang et al. · tsinghua

Deep neural networks (DNNs) have demonstrated their superiority in practice. Arguably, the rapid development of DNNs is largely benefited from high-quality (open-sourced) datasets, based on which researchers and developers can easily evaluate and improve their learning methods. Since the data collection is usually time-consuming or even expensive, how to protect their copyrights is of great significance and worth further exploration. In this paper, we revisit dataset ownership verification. We find that existing verification methods introduced new security risks in DNNs trained on the protected dataset, due to the targeted nature of poison-only backdoor watermarks. To alleviate this problem, in this work, we explore the untargeted backdoor watermarking scheme, where the abnormal model behaviors are not deterministic. Specifically, we introduce two dispersibilities and prove their correlation, based on which we design the untargeted backdoor watermark under both poisoned-label and clean-label settings. We also discuss how to use the proposed untargeted backdoor watermark for dataset ownership verification. Experiments on benchmark datasets verify the effectiveness of our methods and their resistance to existing backdoor defenses. Our codes are available at \url{https://github.com/THUYimingLi/Untargeted_Backdoor_Watermark}.

Zhangheng Li, Tianlong Chen, Linyi Li et al.

With the rapid development of deep learning, the sizes of neural networks become larger and larger so that the training and inference often overwhelm the hardware resources. Given the fact that neural networks are often over-parameterized, one effective way to reduce such computational overhead is neural network pruning, by removing redundant parameters from trained neural networks. It has been recently observed that pruning can not only reduce computational overhead but also can improve empirical robustness of deep neural networks (NNs), potentially owing to removing spurious correlations while preserving the predictive accuracies. This paper for the first time demonstrates that pruning can generally improve certified robustness for ReLU-based NNs under the complete verification setting. Using the popular Branch-and-Bound (BaB) framework, we find that pruning can enhance the estimated bound tightness of certified robustness verification, by alleviating linear relaxation and sub-domain split problems. We empirically verify our findings with off-the-shelf pruning methods and further present a new stability-based pruning method tailored for reducing neuron instability, that outperforms existing pruning methods in enhancing certified robustness. Our experiments show that by appropriately pruning an NN, its certified accuracy can be boosted up to 8.2% under standard training, and up to 24.5% under adversarial training on the CIFAR10 dataset. We additionally observe the existence of certified lottery tickets that can match both standard and certified robust accuracies of the original dense models across different datasets. Our findings offer a new angle to study the intriguing interaction between sparsity and robustness, i.e. interpreting the interaction of sparsity and certified robustness via neuron stability. Codes are available at: https://github.com/VITA-Group/CertifiedPruning.

Fengqing Jiang, Zhangchen Xu, Luyao Niu et al. · uw

Large language models (LLMs) are increasingly deployed as the service backend for LLM-integrated applications such as code completion and AI-powered search. LLM-integrated applications serve as middleware to refine users' queries with domain-specific knowledge to better inform LLMs and enhance the responses. Despite numerous opportunities and benefits, LLM-integrated applications also introduce new attack surfaces. Understanding, minimizing, and eliminating these emerging attack surfaces is a new area of research. In this work, we consider a setup where the user and LLM interact via an LLM-integrated application in the middle. We focus on the communication rounds that begin with user's queries and end with LLM-integrated application returning responses to the queries, powered by LLMs at the service backend. For this query-response protocol, we identify potential vulnerabilities that can originate from the malicious application developer or from an outsider threat initiator that is able to control the database access, manipulate and poison data that are high-risk for the user. Successful exploits of the identified vulnerabilities result in the users receiving responses tailored to the intent of a threat initiator. We assess such threats against LLM-integrated applications empowered by OpenAI GPT-3.5 and GPT-4. Our empirical results show that the threats can effectively bypass the restrictions and moderation policies of OpenAI, resulting in users receiving responses that contain bias, toxic content, privacy risk, and disinformation. To mitigate those threats, we identify and define four key properties, namely integrity, source identification, attack detectability, and utility preservation, that need to be satisfied by a safe LLM-integrated application. Based on these properties, we develop a lightweight, threat-agnostic defense that mitigates both insider and outsider threats.

Mengdi Xu, Zuxin Liu, Peide Huang et al. · cmu

A trustworthy reinforcement learning algorithm should be competent in solving challenging real-world problems, including {robustly} handling uncertainties, satisfying {safety} constraints to avoid catastrophic failures, and {generalizing} to unseen scenarios during deployments. This study aims to overview these main perspectives of trustworthy reinforcement learning considering its intrinsic vulnerabilities on robustness, safety, and generalizability. In particular, we give rigorous formulations, categorize corresponding methodologies, and discuss benchmarks for each perspective. Moreover, we provide an outlook section to spur promising future directions with a brief discussion on extrinsic vulnerabilities considering human feedback. We hope this survey could bring together separate threads of studies together in a unified framework and promote the trustworthiness of reinforcement learning.

Mansur Arief, Zhepeng Cen, Zhenyuan Liu et al. · cmu

Evaluating the performance of autonomous vehicles (AV) and their complex subsystems to high precision under naturalistic circumstances remains a challenge, especially when failure or dangerous cases are rare. Rarity does not only require an enormous sample size for a naive method to achieve high confidence estimation, but it also causes dangerous underestimation of the true failure rate and it is extremely hard to detect. Meanwhile, the state-of-the-art approach that comes with a correctness guarantee can only compute an upper bound for the failure rate under certain conditions, which could limit its practical uses. In this work, we present Deep Importance Sampling (Deep IS) framework that utilizes a deep neural network to obtain an efficient IS that is on par with the state-of-the-art, capable of reducing the required sample size 43 times smaller than the naive sampling method to achieve 10% relative error and while producing an estimate that is much less conservative. Our high-dimensional experiment estimating the misclassification rate of one of the state-of-the-art traffic sign classifiers further reveals that this efficiency still holds true even when the target is very small, achieving over 600 times efficiency boost. This highlights the potential of Deep IS in providing a precise estimate even against high-dimensional uncertainties.

Junfeng Guo, Yiming Li, Lixu Wang et al.

The prosperity of deep neural networks (DNNs) is largely benefited from open-source datasets, based on which users can evaluate and improve their methods. In this paper, we revisit backdoor-based dataset ownership verification (DOV), which is currently the only feasible approach to protect the copyright of open-source datasets. We reveal that these methods are fundamentally harmful given that they could introduce malicious misclassification behaviors to watermarked DNNs by the adversaries. In this paper, we design DOV from another perspective by making watermarked models (trained on the protected dataset) correctly classify some `hard' samples that will be misclassified by the benign model. Our method is inspired by the generalization property of DNNs, where we find a \emph{hardly-generalized domain} for the original dataset (as its \emph{domain watermark}). It can be easily learned with the protected dataset containing modified samples. Specifically, we formulate the domain generation as a bi-level optimization and propose to optimize a set of visually-indistinguishable clean-label modified data with similar effects to domain-watermarked samples from the hardly-generalized domain to ensure watermark stealthiness. We also design a hypothesis-test-guided ownership verification via our domain watermark and provide the theoretical analyses of our method. Extensive experiments on three benchmark datasets are conducted, which verify the effectiveness of our method and its resistance to potential adaptive methods. The code for reproducing main experiments is available at \url{https://github.com/JunfengGo/Domain-Watermark}.

Chaowei Xiao, Zhongzhu Chen, Kun Jin et al.

Diffusion models have been recently employed to improve certified robustness through the process of denoising. However, the theoretical understanding of why diffusion models are able to improve the certified robustness is still lacking, preventing from further improvement. In this study, we close this gap by analyzing the fundamental properties of diffusion models and establishing the conditions under which they can enhance certified robustness. This deeper understanding allows us to propose a new method DensePure, designed to improve the certified robustness of a pretrained model (i.e. classifier). Given an (adversarial) input, DensePure consists of multiple runs of denoising via the reverse process of the diffusion model (with different random seeds) to get multiple reversed samples, which are then passed through the classifier, followed by majority voting of inferred labels to make the final prediction. This design of using multiple runs of denoising is informed by our theoretical analysis of the conditional distribution of the reversed sample. Specifically, when the data density of a clean sample is high, its conditional density under the reverse process in a diffusion model is also high; thus sampling from the latter conditional distribution can purify the adversarial example and return the corresponding clean sample with a high probability. By using the highest density point in the conditional distribution as the reversed sample, we identify the robust region of a given instance under the diffusion model's reverse process. We show that this robust region is a union of multiple convex sets, and is potentially much larger than the robust regions identified in previous works. In practice, DensePure can approximate the label of the high density region in the conditional distribution so that it can enhance certified robustness.

Zhaoyu Chen, Bo Li, Jianghe Xu et al.

Patch attacks, one of the most threatening forms of physical attack in adversarial examples, can lead networks to induce misclassification by modifying pixels arbitrarily in a continuous region. Certifiable patch defense can guarantee robustness that the classifier is not affected by patch attacks. Existing certifiable patch defenses sacrifice the clean accuracy of classifiers and only obtain a low certified accuracy on toy datasets. Furthermore, the clean and certified accuracy of these methods is still significantly lower than the accuracy of normal classification networks, which limits their application in practice. To move towards a practical certifiable patch defense, we introduce Vision Transformer (ViT) into the framework of Derandomized Smoothing (DS). Specifically, we propose a progressive smoothed image modeling task to train Vision Transformer, which can capture the more discriminable local context of an image while preserving the global semantic information. For efficient inference and deployment in the real world, we innovatively reconstruct the global self-attention structure of the original ViT into isolated band unit self-attention. On ImageNet, under 2% area patch attacks our method achieves 41.70% certified accuracy, a nearly 1-fold increase over the previous best method (26.00%). Simultaneously, our method achieves 78.58% clean accuracy, which is quite close to the normal ResNet-101 accuracy. Extensive experiments show that our method obtains state-of-the-art clean and certified accuracy with inferring efficiently on CIFAR-10 and ImageNet.

Jie Zhang, Bo Li, Chen Chen et al.

In Federated Learning (FL), models are as fragile as centrally trained models against adversarial examples. However, the adversarial robustness of federated learning remains largely unexplored. This paper casts light on the challenge of adversarial robustness of federated learning. To facilitate a better understanding of the adversarial vulnerability of the existing FL methods, we conduct comprehensive robustness evaluations on various attacks and adversarial training methods. Moreover, we reveal the negative impacts induced by directly adopting adversarial training in FL, which seriously hurts the test accuracy, especially in non-IID settings. In this work, we propose a novel algorithm called Decision Boundary based Federated Adversarial Training (DBFAT), which consists of two components (local re-weighting and global regularization) to improve both accuracy and robustness of FL systems. Extensive experiments on multiple datasets demonstrate that DBFAT consistently outperforms other baselines under both IID and non-IID settings.

Zhaoyu Chen, Bo Li, Shuang Wu et al.

Deep neural networks (DNNs) have been showed to be highly vulnerable to imperceptible adversarial perturbations. As a complementary type of adversary, patch attacks that introduce perceptible perturbations to the images have attracted the interest of researchers. Existing patch attacks rely on the architecture of the model or the probabilities of predictions and perform poorly in the decision-based setting, which can still construct a perturbation with the minimal information exposed -- the top-1 predicted label. In this work, we first explore the decision-based patch attack. To enhance the attack efficiency, we model the patches using paired key-points and use targeted images as the initialization of patches, and parameter optimizations are all performed on the integer domain. Then, we propose a differential evolutionary algorithm named DevoPatch for query-efficient decision-based patch attacks. Experiments demonstrate that DevoPatch outperforms the state-of-the-art black-box patch attacks in terms of patch area and attack success rate within a given query budget on image classification and face verification. Additionally, we conduct the vulnerability evaluation of ViT and MLP on image classification in the decision-based patch attack setting for the first time. Using DevoPatch, we can evaluate the robustness of models to black-box patch attacks. We believe this method could inspire the design and deployment of robust vision models based on various DNN architectures in the future.

Xiaojun Xu, Linyi Li, Bo Li

Recent studies show that training deep neural networks (DNNs) with Lipschitz constraints are able to enhance adversarial robustness and other model properties such as stability. In this paper, we propose a layer-wise orthogonal training method (LOT) to effectively train 1-Lipschitz convolution layers via parametrizing an orthogonal matrix with an unconstrained matrix. We then efficiently compute the inverse square root of a convolution kernel by transforming the input domain to the Fourier frequency domain. On the other hand, as existing works show that semi-supervised training helps improve empirical robustness, we aim to bridge the gap and prove that semi-supervised learning also improves the certified robustness of Lipschitz-bounded models. We conduct comprehensive evaluations for LOT under different settings. We show that LOT significantly outperforms baselines regarding deterministic l2 certified robustness, and scales to deeper neural networks. Under the supervised scenario, we improve the state-of-the-art certified robustness for all architectures (e.g. from 59.04% to 63.50% on CIFAR-10 and from 32.57% to 34.59% on CIFAR-100 at radius rho = 36/255 for 40-layer networks). With semi-supervised learning over unlabelled data, we are able to improve state-of-the-art certified robustness on CIFAR-10 at rho = 108/255 from 36.04% to 42.39%. In addition, LOT consistently outperforms baselines on different model architectures with only 1/3 evaluation time.

Mantas Mazeika, Bo Li, David Forsyth

Model stealing attacks present a dilemma for public machine learning APIs. To protect financial investments, companies may be forced to withhold important information about their models that could facilitate theft, including uncertainty estimates and prediction explanations. This compromise is harmful not only to users but also to external transparency. Model stealing defenses seek to resolve this dilemma by making models harder to steal while preserving utility for benign users. However, existing defenses have poor performance in practice, either requiring enormous computational overheads or severe utility trade-offs. To meet these challenges, we present a new approach to model stealing defenses called gradient redirection. At the core of our approach is a provably optimal, efficient algorithm for steering an adversary's training updates in a targeted manner. Combined with improvements to surrogate networks and a novel coordinated defense strategy, our gradient redirection defense, called GRAD${}^2$, achieves small utility trade-offs and low computational overhead, outperforming the best prior defenses. Moreover, we demonstrate how gradient redirection enables reprogramming the adversary with arbitrary behavior, which we hope will foster work on new avenues of defense.

Kaixun Jiang, Zhaoyu Chen, Hao Huang et al.

Although Deep Neural Networks (DNNs) have demonstrated excellent performance, they are vulnerable to adversarial patches that introduce perceptible and localized perturbations to the input. Generating adversarial patches on images has received much attention, while adversarial patches on videos have not been well investigated. Further, decision-based attacks, where attackers only access the predicted hard labels by querying threat models, have not been well explored on video models either, even if they are practical in real-world video recognition scenes. The absence of such studies leads to a huge gap in the robustness assessment for video models. To bridge this gap, this work first explores decision-based patch attacks on video models. We analyze that the huge parameter space brought by videos and the minimal information returned by decision-based models both greatly increase the attack difficulty and query burden. To achieve a query-efficient attack, we propose a spatial-temporal differential evolution (STDE) framework. First, STDE introduces target videos as patch textures and only adds patches on keyframes that are adaptively selected by temporal difference. Second, STDE takes minimizing the patch area as the optimization objective and adopts spatialtemporal mutation and crossover to search for the global optimum without falling into the local optimum. Experiments show STDE has demonstrated state-of-the-art performance in terms of threat, efficiency and imperceptibility. Hence, STDE has the potential to be a powerful tool for evaluating the robustness of video recognition models.

Jiawei Zhang, Linyi Li, Ce Zhang et al.

Despite great recent advances achieved by deep neural networks (DNNs), they are often vulnerable to adversarial attacks. Intensive research efforts have been made to improve the robustness of DNNs; however, most empirical defenses can be adaptively attacked again, and the theoretically certified robustness is limited, especially on large-scale datasets. One potential root cause of such vulnerabilities for DNNs is that although they have demonstrated powerful expressiveness, they lack the reasoning ability to make robust and reliable predictions. In this paper, we aim to integrate domain knowledge to enable robust learning with the reasoning paradigm. In particular, we propose a certifiably robust learning with reasoning pipeline (CARE), which consists of a learning component and a reasoning component. Concretely, we use a set of standard DNNs to serve as the learning component to make semantic predictions, and we leverage the probabilistic graphical models, such as Markov logic networks (MLN), to serve as the reasoning component to enable knowledge/logic reasoning. However, it is known that the exact inference of MLN (reasoning) is #P-complete, which limits the scalability of the pipeline. To this end, we propose to approximate the MLN inference via variational inference based on an efficient expectation maximization algorithm. In particular, we leverage graph convolutional networks (GCNs) to encode the posterior distribution during variational inference and update the parameters of GCNs (E-step) and the weights of knowledge rules in MLN (M-step) iteratively. We conduct extensive experiments on different datasets and show that CARE achieves significantly higher certified robustness compared with the state-of-the-art baselines. We additionally conducted different ablation studies to demonstrate the empirical robustness of CARE and the effectiveness of different knowledge integration.

Fan Wu, Linyi Li, Chejian Xu et al.

As reinforcement learning (RL) has achieved near human-level performance in a variety of tasks, its robustness has raised great attention. While a vast body of research has explored test-time (evasion) attacks in RL and corresponding defenses, its robustness against training-time (poisoning) attacks remains largely unanswered. In this work, we focus on certifying the robustness of offline RL in the presence of poisoning attacks, where a subset of training trajectories could be arbitrarily manipulated. We propose the first certification framework, COPA, to certify the number of poisoning trajectories that can be tolerated regarding different certification criteria. Given the complex structure of RL, we propose two certification criteria: per-state action stability and cumulative reward bound. To further improve the certification, we propose new partition and aggregation protocols to train robust policies. We further prove that some of the proposed certification methods are theoretically tight and some are NP-Complete problems. We leverage COPA to certify three RL environments trained with different algorithms and conclude: (1) The proposed robust aggregation protocols such as temporal aggregation can significantly improve the certifications; (2) Our certification for both per-state action stability and cumulative reward bound are efficient and tight; (3) The certification for different training algorithms and environments are different, implying their intrinsic robustness properties. All experimental results are available at https://copa-leaderboard.github.io.

Ayoosh Bansal, Hunmin Kim, Simon Yu et al.

Advances in deep learning have revolutionized cyber-physical applications, including the development of Autonomous Vehicles. However, real-world collisions involving autonomous control of vehicles have raised significant safety concerns regarding the use of Deep Neural Networks (DNN) in safety-critical tasks, particularly Perception. The inherent unverifiability of DNNs poses a key challenge in ensuring their safe and reliable operation. In this work, we propose Perception Simplex (PS), a fault-tolerant application architecture designed for obstacle detection and collision avoidance. We analyze an existing LiDAR-based classical obstacle detection algorithm to establish strict bounds on its capabilities and limitations. Such analysis and verification have not been possible for deep learning-based perception systems yet. By employing verifiable obstacle detection algorithms, PS identifies obstacle existence detection faults in the output of unverifiable DNN-based object detectors. When faults with potential collision risks are detected, appropriate corrective actions are initiated. Through extensive analysis and software-in-the-loop simulations, we demonstrate that PS provides predictable and deterministic fault tolerance against obstacle existence detection faults, establishing a robust safety guarantee.

Dinuka Sahabandu, Arezoo Rajabi, Luyao Niu et al.

Machine learning models in the wild have been shown to be vulnerable to Trojan attacks during training. Although many detection mechanisms have been proposed, strong adaptive attackers have been shown to be effective against them. In this paper, we aim to answer the questions considering an intelligent and adaptive adversary: (i) What is the minimal amount of instances required to be Trojaned by a strong attacker? and (ii) Is it possible for such an attacker to bypass strong detection mechanisms? We provide an analytical characterization of adversarial capability and strategic interactions between the adversary and detection mechanism that take place in such models. We characterize adversary capability in terms of the fraction of the input dataset that can be embedded with a Trojan trigger. We show that the loss function has a submodular structure, which leads to the design of computationally efficient algorithms to determine this fraction with provable bounds on optimality. We propose a Submodular Trojan algorithm to determine the minimal fraction of samples to inject a Trojan trigger. To evade detection of the Trojaned model, we model strategic interactions between the adversary and Trojan detection mechanism as a two-player game. We show that the adversary wins the game with probability one, thus bypassing detection. We establish this by proving that output probability distributions of a Trojan model and a clean model are identical when following the Min-Max (MM) Trojan algorithm. We perform extensive evaluations of our algorithms on MNIST, CIFAR-10, and EuroSAT datasets. The results show that (i) with Submodular Trojan algorithm, the adversary needs to embed a Trojan trigger into a very small fraction of samples to achieve high accuracy on both Trojan and clean samples, and (ii) the MM Trojan algorithm yields a trained Trojan model that evades detection with probability 1.

Yinyi Luo, Yiqiao Jin, Weichen Yu et al.

While large language model (LLM) multi-agent systems achieve superior reasoning performance through iterative debate, practical deployment is limited by their high computational cost and error propagation. This paper proposes AgentArk, a novel framework to distill multi-agent dynamics into the weights of a single model, effectively transforming explicit test-time interactions into implicit model capabilities. This equips a single agent with the intelligence of multi-agent systems while remaining computationally efficient. Specifically, we investigate three hierarchical distillation strategies across various models, tasks, scaling, and scenarios: reasoning-enhanced fine-tuning; trajectory-based augmentation; and process-aware distillation. By shifting the burden of computation from inference to training, the distilled models preserve the efficiency of one agent while exhibiting strong reasoning and self-correction performance of multiple agents. They further demonstrate enhanced robustness and generalization across diverse reasoning tasks. We hope this work can shed light on future research on efficient and robust multi-agent development. Our code is at https://github.com/AIFrontierLab/AgentArk.

Kaixun Jiang, Zhaoyu Chen, Haijing Guo et al.

Preference alignment in diffusion models has primarily focused on benign human preferences (e.g., aesthetic). In this paper, we propose a novel perspective: framing unrestricted adversarial example generation as a problem of aligning with adversary preferences. Unlike benign alignment, adversarial alignment involves two inherently conflicting preferences: visual consistency and attack effectiveness, which often lead to unstable optimization and reward hacking (e.g., reducing visual quality to improve attack success). To address this, we propose APA (Adversary Preferences Alignment), a two-stage framework that decouples conflicting preferences and optimizes each with differentiable rewards. In the first stage, APA fine-tunes LoRA to improve visual consistency using rule-based similarity reward. In the second stage, APA updates either the image latent or prompt embedding based on feedback from a substitute classifier, guided by trajectory-level and step-wise rewards. To enhance black-box transferability, we further incorporate a diffusion augmentation strategy. Experiments demonstrate that APA achieves significantly better attack transferability while maintaining high visual consistency, inspiring further research to approach adversarial attacks from an alignment perspective. Code will be available at https://github.com/deep-kaixun/APA.

Yige Li, Xixiang Lyu, Nodens Koren et al.

Backdoor attack has emerged as a major security threat to deep neural networks (DNNs). While existing defense methods have demonstrated promising results on detecting or erasing backdoors, it is still not clear whether robust training methods can be devised to prevent the backdoor triggers being injected into the trained model in the first place. In this paper, we introduce the concept of \emph{anti-backdoor learning}, aiming to train \emph{clean} models given backdoor-poisoned data. We frame the overall learning process as a dual-task of learning the \emph{clean} and the \emph{backdoor} portions of data. From this view, we identify two inherent characteristics of backdoor attacks as their weaknesses: 1) the models learn backdoored data much faster than learning with clean data, and the stronger the attack the faster the model converges on backdoored data; 2) the backdoor task is tied to a specific class (the backdoor target class). Based on these two weaknesses, we propose a general learning scheme, Anti-Backdoor Learning (ABL), to automatically prevent backdoor attacks during training. ABL introduces a two-stage \emph{gradient ascent} mechanism for standard training to 1) help isolate backdoor examples at an early training stage, and 2) break the correlation between backdoor examples and the target class at a later training stage. Through extensive experiments on multiple benchmark datasets against 10 state-of-the-art attacks, we empirically show that ABL-trained models on backdoor-poisoned data achieve the same performance as they were trained on purely clean data. Code is available at \url{https://github.com/bboylyg/ABL}.

Yulong Cao*, Ningfei Wang*, Chaowei Xiao* et al.

In Autonomous Driving (AD) systems, perception is both security and safety critical. Despite various prior studies on its security issues, all of them only consider attacks on camera- or LiDAR-based AD perception alone. However, production AD systems today predominantly adopt a Multi-Sensor Fusion (MSF) based design, which in principle can be more robust against these attacks under the assumption that not all fusion sources are (or can be) attacked at the same time. In this paper, we present the first study of security issues of MSF-based perception in AD systems. We directly challenge the basic MSF design assumption above by exploring the possibility of attacking all fusion sources simultaneously. This allows us for the first time to understand how much security guarantee MSF can fundamentally provide as a general defense strategy for AD perception. We formulate the attack as an optimization problem to generate a physically-realizable, adversarial 3D-printed object that misleads an AD system to fail in detecting it and thus crash into it. We propose a novel attack pipeline that addresses two main design challenges: (1) non-differentiable target camera and LiDAR sensing systems, and (2) non-differentiable cell-level aggregated features popularly used in LiDAR-based AD perception. We evaluate our attack on MSF included in representative open-source industry-grade AD systems in real-world driving scenarios. Our results show that the attack achieves over 90% success rate across different object types and MSF. Our attack is also found stealthy, robust to victim positions, transferable across MSF algorithms, and physical-world realizable after being 3D-printed and captured by LiDAR and camera devices. To concretely assess the end-to-end safety impact, we further perform simulation evaluation and show that it can cause a 100% vehicle collision rate for an industry-grade AD system.

Linyi Li, Tao Xie, Bo Li

Great advances in deep neural networks (DNNs) have led to state-of-the-art performance on a wide range of tasks. However, recent studies have shown that DNNs are vulnerable to adversarial attacks, which have brought great concerns when deploying these models to safety-critical applications such as autonomous driving. Different defense approaches have been proposed against adversarial attacks, including: a) empirical defenses, which can usually be adaptively attacked again without providing robustness certification; and b) certifiably robust approaches, which consist of robustness verification providing the lower bound of robust accuracy against any attacks under certain conditions and corresponding robust training approaches. In this paper, we systematize certifiably robust approaches and related practical and theoretical implications and findings. We also provide the first comprehensive benchmark on existing robustness verification and training approaches on different datasets. In particular, we 1) provide a taxonomy for the robustness verification and training approaches, as well as summarize the methodologies for representative algorithms, 2) reveal the characteristics, strengths, limitations, and fundamental connections among these approaches, 3) discuss current research progresses, theoretical barriers, main challenges, and future directions for certifiably robust approaches for DNNs, and 4) provide an open-sourced unified platform to evaluate 20+ representative certifiably robust approaches.

Huan Zhang, Hongge Chen, Chaowei Xiao et al.

A deep reinforcement learning (DRL) agent observes its states through observations, which may contain natural measurement errors or adversarial noises. Since the observations deviate from the true states, they can mislead the agent into making suboptimal actions. Several works have shown this vulnerability via adversarial attacks, but existing approaches on improving the robustness of DRL under this setting have limited success and lack for theoretical principles. We show that naively applying existing techniques on improving robustness for classification tasks, like adversarial training, is ineffective for many RL tasks. We propose the state-adversarial Markov decision process (SA-MDP) to study the fundamental properties of this problem, and develop a theoretically principled policy regularization which can be applied to a large family of DRL algorithms, including proximal policy optimization (PPO), deep deterministic policy gradient (DDPG) and deep Q networks (DQN), for both discrete and continuous action control problems. We significantly improve the robustness of PPO, DDPG and DQN agents under a suite of strong white box adversarial attacks, including new attacks of our own. Additionally, we find that a robust policy noticeably improves DRL performance even without an adversary in a number of environments. Our code is available at https://github.com/chenhongge/StateAdvDRL.

Bertie Vidgen, Adarsh Agrawal, Ahmed M. Ahmed et al. · deepmind, oxford

This paper introduces v0.5 of the AI Safety Benchmark, which has been created by the MLCommons AI Safety Working Group. The AI Safety Benchmark has been designed to assess the safety risks of AI systems that use chat-tuned language models. We introduce a principled approach to specifying and constructing the benchmark, which for v0.5 covers only a single use case (an adult chatting to a general-purpose assistant in English), and a limited set of personas (i.e., typical users, malicious users, and vulnerable users). We created a new taxonomy of 13 hazard categories, of which 7 have tests in the v0.5 benchmark. We plan to release version 1.0 of the AI Safety Benchmark by the end of 2024. The v1.0 benchmark will provide meaningful insights into the safety of AI systems. However, the v0.5 benchmark should not be used to assess the safety of AI systems. We have sought to fully document the limitations, flaws, and challenges of v0.5. This release of v0.5 of the AI Safety Benchmark includes (1) a principled approach to specifying and constructing the benchmark, which comprises use cases, types of systems under test (SUTs), language and context, personas, tests, and test items; (2) a taxonomy of 13 hazard categories with definitions and subcategories; (3) tests for seven of the hazard categories, each comprising a unique set of test items, i.e., prompts. There are 43,090 test items in total, which we created with templates; (4) a grading system for AI systems against the benchmark; (5) an openly available platform, and downloadable tool, called ModelBench that can be used to evaluate the safety of AI systems on the benchmark; (6) an example evaluation report which benchmarks the performance of over a dozen openly available chat-tuned language models; (7) a test specification for the benchmark.

Zijian Huang, Wenda Chu, Linyi Li et al.

Multi-sensor fusion systems (MSFs) play a vital role as the perception module in modern autonomous vehicles (AVs). Therefore, ensuring their robustness against common and realistic adversarial semantic transformations, such as rotation and shifting in the physical world, is crucial for the safety of AVs. While empirical evidence suggests that MSFs exhibit improved robustness compared to single-modal models, they are still vulnerable to adversarial semantic transformations. Despite the proposal of empirical defenses, several works show that these defenses can be attacked again by new adaptive attacks. So far, there is no certified defense proposed for MSFs. In this work, we propose the first robustness certification framework COMMIT certify robustness of multi-sensor fusion systems against semantic attacks. In particular, we propose a practical anisotropic noise mechanism that leverages randomized smoothing with multi-modal data and performs a grid-based splitting method to characterize complex semantic transformations. We also propose efficient algorithms to compute the certification in terms of object detection accuracy and IoU for large-scale MSF models. Empirically, we evaluate the efficacy of COMMIT in different settings and provide a comprehensive benchmark of certified robustness for different MSF models using the CARLA simulation platform. We show that the certification for MSF models is at most 48.39% higher than that of single-modal models, which validates the advantages of MSF models. We believe our certification framework and benchmark will contribute an important step towards certifiably robust AVs in practice.

Dinuka Sahabandu, Xiaojun Xu, Arezoo Rajabi et al.

We propose and analyze an adaptive adversary that can retrain a Trojaned DNN and is also aware of SOTA output-based Trojaned model detectors. We show that such an adversary can ensure (1) high accuracy on both trigger-embedded and clean samples and (2) bypass detection. Our approach is based on an observation that the high dimensionality of the DNN parameters provides sufficient degrees of freedom to simultaneously achieve these objectives. We also enable SOTA detectors to be adaptive by allowing retraining to recalibrate their parameters, thus modeling a co-evolution of parameters of a Trojaned model and detectors. We then show that this co-evolution can be modeled as an iterative game, and prove that the resulting (optimal) solution of this interactive game leads to the adversary successfully achieving the above objectives. In addition, we provide a greedy algorithm for the adversary to select a minimum number of input samples for embedding triggers. We show that for cross-entropy or log-likelihood loss functions used by the DNNs, the greedy algorithm provides provable guarantees on the needed number of trigger-embedded input samples. Extensive experiments on four diverse datasets -- MNIST, CIFAR-10, CIFAR-100, and SpeechCommand -- reveal that the adversary effectively evades four SOTA output-based Trojaned model detectors: MNTD, NeuralCleanse, STRIP, and TABOR.

Zhaoyu Chen, Haijing Guo, Kaixun Jiang et al.

Deep neural networks are vulnerable to adversarial examples that exhibit transferability across various models. Numerous approaches are proposed to enhance the transferability of adversarial examples, including advanced optimization, data augmentation, and model modifications. However, these methods still show limited transferability, particularly in cross-architecture scenarios, such as from CNN to ViT. To achieve high transferability, we propose a technique termed Spatial Adversarial Alignment (SAA), which employs an alignment loss and leverages a witness model to fine-tune the surrogate model. Specifically, SAA consists of two key parts: spatial-aware alignment and adversarial-aware alignment. First, we minimize the divergences of features between the two models in both global and local regions, facilitating spatial alignment. Second, we introduce a self-adversarial strategy that leverages adversarial examples to impose further constraints, aligning features from an adversarial perspective. Through this alignment, the surrogate model is trained to concentrate on the common features extracted by the witness model. This facilitates adversarial attacks on these shared features, thereby yielding perturbations that exhibit enhanced transferability. Extensive experiments on various architectures on ImageNet show that aligned surrogate models based on SAA can provide higher transferable adversarial examples, especially in cross-architecture attacks.

Zhen Xiang, Zidi Xiong, Bo Li

Backdoor (Trojan) attack is a common threat to deep neural networks, where samples from one or more source classes embedded with a backdoor trigger will be misclassified to adversarial target classes. Existing methods for detecting whether a classifier is backdoor attacked are mostly designed for attacks with a single adversarial target (e.g., all-to-one attack). To the best of our knowledge, without supervision, no existing methods can effectively address the more general X2X attack with an arbitrary number of source classes, each paired with an arbitrary target class. In this paper, we propose UMD, the first Unsupervised Model Detection method that effectively detects X2X backdoor attacks via a joint inference of the adversarial (source, target) class pairs. In particular, we first define a novel transferability statistic to measure and select a subset of putative backdoor class pairs based on a proposed clustering approach. Then, these selected class pairs are jointly assessed based on an aggregation of their reverse-engineered trigger size for detection inference, using a robust and unsupervised anomaly detector we proposed. We conduct comprehensive evaluations on CIFAR-10, GTSRB, and Imagenette dataset, and show that our unsupervised UMD outperforms SOTA detectors (even with supervision) by 17%, 4%, and 8%, respectively, in terms of the detection accuracy against diverse X2X attacks. We also show the strong detection performance of UMD against several strong adaptive attacks.

Zhaoyu Chen, Bo Li, Shuang Wu et al.

Unrestricted adversarial attacks typically manipulate the semantic content of an image (e.g., color or texture) to create adversarial examples that are both effective and photorealistic, demonstrating their ability to deceive human perception and deep neural networks with stealth and success. However, current works usually sacrifice unrestricted degrees and subjectively select some image content to guarantee the photorealism of unrestricted adversarial examples, which limits its attack performance. To ensure the photorealism of adversarial examples and boost attack performance, we propose a novel unrestricted attack framework called Content-based Unrestricted Adversarial Attack. By leveraging a low-dimensional manifold that represents natural images, we map the images onto the manifold and optimize them along its adversarial direction. Therefore, within this framework, we implement Adversarial Content Attack based on Stable Diffusion and can generate high transferable unrestricted adversarial examples with various adversarial contents. Extensive experimentation and visualization demonstrate the efficacy of ACA, particularly in surpassing state-of-the-art attacks by an average of 13.3-50.4% and 16.8-48.0% in normally trained models and defense methods, respectively.

Xiaojun Xu, Jacky Yibo Zhang, Evelyn Ma et al.

Machine learning (ML) robustness and domain generalization are fundamentally correlated: they essentially concern data distribution shifts under adversarial and natural settings, respectively. On one hand, recent studies show that more robust (adversarially trained) models are more generalizable. On the other hand, there is a lack of theoretical understanding of their fundamental connections. In this paper, we explore the relationship between regularization and domain transferability considering different factors such as norm regularization and data augmentations (DA). We propose a general theoretical framework proving that factors involving the model function class regularization are sufficient conditions for relative domain transferability. Our analysis implies that ``robustness" is neither necessary nor sufficient for transferability; rather, regularization is a more fundamental perspective for understanding domain transferability. We then discuss popular DA protocols (including adversarial training) and show when they can be viewed as the function class regularization under certain conditions and therefore improve generalization. We conduct extensive experiments to verify our theoretical findings and show several counterexamples where robustness and generalization are negatively correlated on different datasets.

Maurice Weber, Linyi Li, Boxin Wang et al.

Certifying the robustness of model performance under bounded data distribution drifts has recently attracted intensive interest under the umbrella of distributional robustness. However, existing techniques either make strong assumptions on the model class and loss functions that can be certified, such as smoothness expressed via Lipschitz continuity of gradients, or require to solve complex optimization problems. As a result, the wider application of these techniques is currently limited by its scalability and flexibility -- these techniques often do not scale to large-scale datasets with modern deep neural networks or cannot handle loss functions which may be non-smooth such as the 0-1 loss. In this paper, we focus on the problem of certifying distributional robustness for blackbox models and bounded loss functions, and propose a novel certification framework based on the Hellinger distance. Our certification technique scales to ImageNet-scale datasets, complex models, and a diverse set of loss functions. We then focus on one specific application enabled by such scalability and flexibility, i.e., certifying out-of-domain generalization for large neural networks and loss functions such as accuracy and AUC. We experimentally validate our certification method on a number of datasets, ranging from ImageNet, where we provide the first non-vacuous certified out-of-domain generalization, to smaller classification tasks where we are able to compare with the state-of-the-art and show that our method performs considerably better.

Jialuo Chen, Jingyi Wang, Tinglan Peng et al.

Deep learning (DL) models, especially those large-scale and high-performance ones, can be very costly to train, demanding a great amount of data and computational resources. Unauthorized reproduction of DL models can lead to copyright infringement and cause huge economic losses to model owners. Existing copyright protection techniques are mostly based on watermarking, which embeds an owner-specified watermark into the model. While being able to provide exact ownership verification, these techniques are 1) invasive, as they need to tamper with the training process, which may affect the utility or introduce new security risks; 2) prone to adaptive attacks that attempt to remove the watermark; and 3) not robust to the emerging model extraction attacks. Latest fingerprinting work, though being non-invasive, also falls short when facing the diverse and ever-growing attack scenarios. In this paper, we propose a novel testing framework for DL copyright protection: DEEPJUDGE. DEEPJUDGE quantitatively tests the similarities between two DL models: a victim model and a suspect model. It leverages a diverse set of testing metrics and test case generation methods to produce a chain of supporting evidence to help determine whether a suspect model is a copy of the victim model. Advantages of DEEPJUDGE include: 1) non-invasive, as it works directly on the model and does not tamper with the training process; 2) efficient, as it only needs a small set of test cases and a quick scan of models; 3) flexible, as it can easily incorporate new metrics or generation methods to obtain more confident judgement; and 4) fairly robust to model extraction and adaptive attacks. We verify the effectiveness of DEEPJUDGE under typical copyright infringement scenarios, including model finetuning, pruning and extraction, via extensive experiments on both image and speech datasets with a variety of model architectures.

Tao Bai, Jun Zhao, Jinlin Zhu et al.

Intelligent Internet of Things (IoT) systems based on deep neural networks (DNNs) have been widely deployed in the real world. However, DNNs are found to be vulnerable to adversarial examples, which raises people's concerns about intelligent IoT systems' reliability and security. Testing and evaluating the robustness of IoT systems becomes necessary and essential. Recently various attacks and strategies have been proposed, but the efficiency problem remains unsolved properly. Existing methods are either computationally extensive or time-consuming, which is not applicable in practice. In this paper, we propose a novel framework called Attack-Inspired GAN (AI-GAN) to generate adversarial examples conditionally. Once trained, it can generate adversarial perturbations efficiently given input images and target classes. We apply AI-GAN on different datasets in white-box settings, black-box settings and targeted models protected by state-of-the-art defenses. Through extensive experiments, AI-GAN achieves high attack success rates, outperforming existing methods, and reduces generation time significantly. Moreover, for the first time, AI-GAN successfully scales to complex datasets e.g. CIFAR-100 and ImageNet, with about $90\%$ success rates among all classes.

Zhuolin Yang, Linyi Li, Xiaojun Xu et al.

Recent studies show that deep neural networks (DNN) are vulnerable to adversarial examples, which aim to mislead DNNs by adding perturbations with small magnitude. To defend against such attacks, both empirical and theoretical defense approaches have been extensively studied for a single ML model. In this work, we aim to analyze and provide the certified robustness for ensemble ML models, together with the sufficient and necessary conditions of robustness for different ensemble protocols. Although ensemble models are shown more robust than a single model empirically; surprisingly, we find that in terms of the certified robustness the standard ensemble models only achieve marginal improvement compared to a single model. Thus, to explore the conditions that guarantee to provide certifiably robust ensemble ML models, we first prove that diversified gradient and large confidence margin are sufficient and necessary conditions for certifiably robust ensemble models under the model-smoothness assumption. We then provide the bounded model-smoothness analysis based on the proposed Ensemble-before-Smoothing strategy. We also prove that an ensemble model can always achieve higher certified robustness than a single base model under mild conditions. Inspired by the theoretical findings, we propose the lightweight Diversity Regularized Training (DRT) to train certifiably robust ensemble ML models. Extensive experiments show that our DRT enhanced ensembles can consistently achieve higher certified robustness than existing single and ensemble ML models, demonstrating the state-of-the-art certified L2-robustness on MNIST, CIFAR-10, and ImageNet datasets.

Nezihe Merve Gürel, Xiangyu Qi, Luka Rimanic et al.

Despite the great successes achieved by deep neural networks (DNNs), recent studies show that they are vulnerable against adversarial examples, which aim to mislead DNNs by adding small adversarial perturbations. Several defenses have been proposed against such attacks, while many of them have been adaptively attacked. In this work, we aim to enhance the ML robustness from a different perspective by leveraging domain knowledge: We propose a Knowledge Enhanced Machine Learning Pipeline (KEMLP) to integrate domain knowledge (i.e., logic relationships among different predictions) into a probabilistic graphical model via first-order logic rules. In particular, we develop KEMLP by integrating a diverse set of weak auxiliary models based on their logical relationships to the main DNN model that performs the target task. Theoretically, we provide convergence results and prove that, under mild conditions, the prediction of KEMLP is more robust than that of the main DNN model. Empirically, we take road sign recognition as an example and leverage the relationships between road signs and their shapes and contents as domain knowledge. We show that compared with adversarial training and other baselines, KEMLP achieves higher robustness against physical attacks, $\mathcal{L}_p$ bounded attacks, unforeseen attacks, and natural corruptions under both whitebox and blackbox settings, while still maintaining high clean accuracy.

Zhuolin Yang, Linyi Li, Xiaojun Xu et al.

Adversarial Transferability is an intriguing property - adversarial perturbation crafted against one model is also effective against another model, while these models are from different model families or training processes. To better protect ML systems against adversarial attacks, several questions are raised: what are the sufficient conditions for adversarial transferability and how to bound it? Is there a way to reduce the adversarial transferability in order to improve the robustness of an ensemble ML model? To answer these questions, in this work we first theoretically analyze and outline sufficient conditions for adversarial transferability between models; then propose a practical algorithm to reduce the transferability between base models within an ensemble to improve its robustness. Our theoretical analysis shows that only promoting the orthogonality between gradients of base models is not enough to ensure low transferability; in the meantime, the model smoothness is an important factor to control the transferability. We also provide the lower and upper bounds of adversarial transferability under certain conditions. Inspired by our theoretical analysis, we propose an effective Transferability Reduced Smooth(TRS) ensemble training strategy to train a robust ensemble with low transferability by enforcing both gradient orthogonality and model smoothness between base models. We conduct extensive experiments on TRS and compare with 6 state-of-the-art ensemble baselines against 8 whitebox attacks on different datasets, demonstrating that the proposed TRS outperforms all baselines significantly.

Shihao Zhao, Xingjun Ma, Yisen Wang et al.

Deep neural networks (DNNs) are increasingly deployed in different applications to achieve state-of-the-art performance. However, they are often applied as a black box with limited understanding of what knowledge the model has learned from the data. In this paper, we focus on image classification and propose a method to visualize and understand the class-wise knowledge (patterns) learned by DNNs under three different settings including natural, backdoor and adversarial. Different to existing visualization methods, our method searches for a single predictive pattern in the pixel space to represent the knowledge learned by the model for each class. Based on the proposed method, we show that DNNs trained on natural (clean) data learn abstract shapes along with some texture, and backdoored models learn a suspicious pattern for the backdoored class. Interestingly, the phenomenon that DNNs can learn a single predictive pattern for each class indicates that DNNs can learn a backdoor even from clean data, and the pattern itself is a backdoor trigger. In the adversarial setting, we show that adversarially trained models tend to learn more simplified shape patterns. Our method can serve as a useful tool to better understand the knowledge learned by DNNs on different datasets under different settings.

Micah Goldblum, Dimitris Tsipras, Chulin Xie et al.

As machine learning systems grow in scale, so do their training data requirements, forcing practitioners to automate and outsource the curation of training data in order to achieve state-of-the-art performance. The absence of trustworthy human supervision over the data collection process exposes organizations to security vulnerabilities; training data can be manipulated to control and degrade the downstream behaviors of learned models. The goal of this work is to systematically categorize and discuss a wide range of dataset vulnerabilities and exploits, approaches for defending against these threats, and an array of open problems in this space. In addition to describing various poisoning and backdoor threat models and the relationships among them, we develop their unified taxonomy.

Bo Li, Yezhen Wang, Shanghang Zhang et al.

The success of supervised learning hinges on the assumption that the training and test data come from the same underlying distribution, which is often not valid in practice due to potential distribution shift. In light of this, most existing methods for unsupervised domain adaptation focus on achieving domain-invariant representations and small source domain error. However, recent works have shown that this is not sufficient to guarantee good generalization on the target domain, and in fact, is provably detrimental under label distribution shift. Furthermore, in many real-world applications it is often feasible to obtain a small amount of labeled data from the target domain and use them to facilitate model training with source data. Inspired by the above observations, in this paper we propose the first method that aims to simultaneously learn invariant representations and risks under the setting of semi-supervised domain adaptation (Semi-DA). First, we provide a finite sample bound for both classification and regression problems under Semi-DA. The bound suggests a principled way to obtain target generalization, i.e. by aligning both the marginal and conditional distributions across domains in feature space. Motivated by this, we then introduce the LIRR algorithm for jointly \textbf{L}earning \textbf{I}nvariant \textbf{R}epresentations and \textbf{R}isks. Finally, extensive experiments are conducted on both classification and regression tasks, which demonstrates LIRR consistently achieves state-of-the-art performance and significant improvements compared with the methods that only learn invariant representations or invariant risks.

Anpeng Wu, Kun Kuang, Junkun Yuan et al.

The fundamental problem in treatment effect estimation from observational data is confounder identification and balancing. Most of the previous methods realized confounder balancing by treating all observed pre-treatment variables as confounders, ignoring further identifying confounders and non-confounders. In general, not all the observed pre-treatment variables are confounders that refer to the common causes of the treatment and the outcome, some variables only contribute to the treatment and some only contribute to the outcome. Balancing those non-confounders, including instrumental variables and adjustment variables, would generate additional bias for treatment effect estimation. By modeling the different causal relations among observed pre-treatment variables, treatment and outcome, we propose a synergistic learning framework to 1) identify confounders by learning decomposed representations of both confounders and non-confounders, 2) balance confounder with sample re-weighting technique, and simultaneously 3) estimate the treatment effect in observational studies via counterfactual inference. Empirical results on synthetic and real-world datasets demonstrate that the proposed method can precisely decompose confounders and achieve a more precise estimation of treatment effect than baselines.

Huichen Li, Xiaojun Xu, Xiaolu Zhang et al.

Machine learning (ML), especially deep neural networks (DNNs) have been widely used in various applications, including several safety-critical ones (e.g. autonomous driving). As a result, recent research about adversarial examples has raised great concerns. Such adversarial attacks can be achieved by adding a small magnitude of perturbation to the input to mislead model prediction. While several whitebox attacks have demonstrated their effectiveness, which assume that the attackers have full access to the machine learning models; blackbox attacks are more realistic in practice. In this paper, we propose a Query-Efficient Boundary-based blackbox Attack (QEBA) based only on model's final prediction labels. We theoretically show why previous boundary-based attack with gradient estimation on the whole gradient space is not efficient in terms of query numbers, and provide optimality analysis for our dimension reduction-based gradient estimation. On the other hand, we conducted extensive experiments on ImageNet and CelebA datasets to evaluate QEBA. We show that compared with the state-of-the-art blackbox attacks, QEBA is able to use a smaller number of queries to achieve a lower magnitude of perturbation with 100% attack success rate. We also show case studies of attacks on real-world APIs including MEGVII Face++ and Microsoft Azure.

Maurice Weber, Xiaojun Xu, Bojan Karlaš et al.

Recent studies have shown that deep neural networks (DNNs) are vulnerable to adversarial attacks, including evasion and backdoor (poisoning) attacks. On the defense side, there have been intensive efforts on improving both empirical and provable robustness against evasion attacks; however, the provable robustness against backdoor attacks still remains largely unexplored. In this paper, we focus on certifying the machine learning model robustness against general threat models, especially backdoor attacks. We first provide a unified framework via randomized smoothing techniques and show how it can be instantiated to certify the robustness against both evasion and backdoor attacks. We then propose the first robust training process, RAB, to smooth the trained model and certify its robustness against backdoor attacks. We prove the robustness bound for machine learning models trained with RAB and prove that our robustness bound is tight. In addition, we theoretically show that it is possible to train the robust smoothed models efficiently for simple models such as K-nearest neighbor classifiers, and we propose an exact smooth-training algorithm that eliminates the need to sample from a noise distribution for such models. Empirically, we conduct comprehensive experiments for different machine learning (ML) models such as DNNs, support vector machines, and K-NN models on MNIST, CIFAR-10, and ImageNette datasets and provide the first benchmark for certified robustness against backdoor attacks. In addition, we evaluate K-NN models on a spambase tabular dataset to demonstrate the advantages of the proposed exact algorithm. Both the theoretic analysis and the comprehensive evaluation on diverse ML models and datasets shed light on further robust learning strategies against general training time attacks.

Linyi Li, Maurice Weber, Xiaojun Xu et al.

As machine learning (ML) systems become pervasive, safeguarding their security is critical. However, recently it has been demonstrated that motivated adversaries are able to mislead ML systems by perturbing test data using semantic transformations. While there exists a rich body of research providing provable robustness guarantees for ML models against $\ell_p$ norm bounded adversarial perturbations, guarantees against semantic perturbations remain largely underexplored. In this paper, we provide TSS -- a unified framework for certifying ML robustness against general adversarial semantic transformations. First, depending on the properties of each transformation, we divide common transformations into two categories, namely resolvable (e.g., Gaussian blur) and differentially resolvable (e.g., rotation) transformations. For the former, we propose transformation-specific randomized smoothing strategies and obtain strong robustness certification. The latter category covers transformations that involve interpolation errors, and we propose a novel approach based on stratified sampling to certify the robustness. Our framework TSS leverages these certification strategies and combines with consistency-enhanced training to provide rigorous certification of robustness. We conduct extensive experiments on over ten types of challenging semantic transformations and show that TSS significantly outperforms the state of the art. Moreover, to the best of our knowledge, TSS is the first approach that achieves nontrivial certified robustness on the large-scale ImageNet dataset. For instance, our framework achieves 30.4% certified robust accuracy against rotation attack (within $\pm 30^\circ$) on ImageNet. Moreover, to consider a broader range of transformations, we show TSS is also robust against adaptive attacks and unforeseen image corruptions such as CIFAR-10-C and ImageNet-C.

Xingxing Wei, Huanqian Yan, Bo Li

Adversarial attacks on video recognition models have been explored recently. However, most existing works treat each video frame equally and ignore their temporal interactions. To overcome this drawback, a few methods try to select some key frames and then perform attacks based on them. Unfortunately, their selection strategy is independent of the attacking step, therefore the resulting performance is limited. Instead, we argue the frame selection phase is closely relevant with the attacking phase. The key frames should be adjusted according to the attacking results. For that, we formulate the black-box video attacks into a Reinforcement Learning (RL) framework. Specifically, the environment in RL is set as the recognition model, and the agent in RL plays the role of frame selecting. By continuously querying the recognition models and receiving the attacking feedback, the agent gradually adjusts its frame selection strategy and adversarial perturbations become smaller and smaller. We conduct a series of experiments with two mainstream video recognition models: C3D and LRCN on the public UCF-101 and HMDB-51 datasets. The results demonstrate that the proposed method can significantly reduce the adversarial perturbations with efficient query times.

Yuheng Zhang, Ruoxi Jia, Hengzhi Pei et al.

This paper studies model-inversion attacks, in which the access to a model is abused to infer information about the training data. Since its first introduction, such attacks have raised serious concerns given that training data usually contain privacy-sensitive information. Thus far, successful model-inversion attacks have only been demonstrated on simple models, such as linear regression and logistic regression. Previous attempts to invert neural networks, even the ones with simple architectures, have failed to produce convincing results. We present a novel attack method, termed the generative model-inversion attack, which can invert deep neural networks with high success rates. Rather than reconstructing private training data from scratch, we leverage partial public information, which can be very generic, to learn a distributional prior via generative adversarial networks (GANs) and use it to guide the inversion process. Moreover, we theoretically prove that a model's predictive power and its vulnerability to inversion attacks are indeed two sides of the same coin---highly predictive models are able to establish a strong correlation between features and labels, which coincides exactly with what an adversary exploits to mount the attacks. Our extensive experiments demonstrate that the proposed attack improves identification accuracy over the existing work by about 75\% for reconstructing face images from a state-of-the-art face recognition classifier. We also show that differential privacy, in its canonical form, is of little avail to defend against our attacks.

Xinlei Pan, Chaowei Xiao, Warren He et al.

Recent studies show that Deep Reinforcement Learning (DRL) models are vulnerable to adversarial attacks, which attack DRL models by adding small perturbations to the observations. However, some attacks assume full availability of the victim model, and some require a huge amount of computation, making them less feasible for real world applications. In this work, we make further explorations of the vulnerabilities of DRL by studying other aspects of attacks on DRL using realistic and efficient attacks. First, we adapt and propose efficient black-box attacks when we do not have access to DRL model parameters. Second, to address the high computational demands of existing attacks, we introduce efficient online sequential attacks that exploit temporal consistency across consecutive steps. Third, we explore the possibility of an attacker perturbing other aspects in the DRL setting, such as the environment dynamics. Finally, to account for imperfections in how an attacker would inject perturbations in the physical world, we devise a method for generating a robust physical perturbations to be printed. The attack is evaluated on a real-world robot under various conditions. We conduct extensive experiments both in simulation such as Atari games, robotics and autonomous driving, and on real-world robotics, to compare the effectiveness of the proposed attacks with baseline approaches. To the best of our knowledge, we are the first to apply adversarial attacks on DRL systems to physical robots.

Bo Li, Tianfu Wu, Lun Zhang et al.

Region-based convolutional neural networks (R-CNN)~\cite{fast_rcnn,faster_rcnn,mask_rcnn} have largely dominated object detection. Operators defined on RoIs (Region of Interests) play an important role in R-CNNs such as RoIPooling~\cite{fast_rcnn} and RoIAlign~\cite{mask_rcnn}. They all only utilize information inside RoIs for RoI prediction, even with their recent deformable extensions~\cite{deformable_cnn}. Although surrounding context is well-known for its importance in object detection, it has yet been integrated in R-CNNs in a flexible and effective way. Inspired by the auto-context work~\cite{auto_context} and the multi-class object layout work~\cite{nms_context}, this paper presents a generic context-mining RoI operator (i.e., \textit{RoICtxMining}) seamlessly integrated in R-CNNs, and the resulting object detection system is termed \textbf{Auto-Context R-CNN} which is trained end-to-end. The proposed RoICtxMining operator is a simple yet effective two-layer extension of the RoIPooling or RoIAlign operator. Centered at an object-RoI, it creates a $3\times 3$ layout to mine contextual information adaptively in the $8$ surrounding context regions on-the-fly. Within each of the $8$ context regions, a context-RoI is mined in term of discriminative power and its RoIPooling / RoIAlign features are concatenated with the object-RoI for final prediction. \textit{The proposed Auto-Context R-CNN is robust to occlusion and small objects, and shows promising vulnerability for adversarial attacks without being adversarially-trained.} In experiments, it is evaluated using RoIPooling as the backbone and shows competitive results on Pascal VOC, Microsoft COCO, and KITTI datasets (including $6.9\%$ mAP improvements over the R-FCN~\cite{rfcn} method on COCO \textit{test-dev} dataset and the first place on both KITTI pedestrian and cyclist detection as of this submission).

Kin Sum Liu, Chaowei Xiao, Bo Li et al.

In this paper we propose a new membership attack method called co-membership attacks against deep generative models including Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs). Specifically, membership attack aims to check whether a given instance x was used in the training data or not. A co-membership attack checks whether the given bundle of n instances were in the training, with the prior knowledge that the bundle was either entirely used in the training or none at all. Successful membership attacks can compromise the privacy of training data when the generative model is published. Our main idea is to cast membership inference of target data x as the optimization of another neural network (called the attacker network) to search for the latent encoding to reproduce x. The final reconstruction error is used directly to conclude whether x was in the training data or not. We conduct extensive experiments on a variety of datasets and generative models showing that: our attacker network outperforms prior membership attacks; co-membership attacks can be substantially more powerful than single attacks; and VAEs are more susceptible to membership attacks compared to GANs.