Fu Song

Zhensu Sun, Xiaoning Du, Fu Song et al.

Code datasets are of immense value for training neural-network-based code completion models, where companies or organizations have made substantial investments to establish and process these datasets. Unluckily, these datasets, either built for proprietary or public usage, face the high risk of unauthorized exploits, resulting from data leakages, license violations, etc. Even worse, the ``black-box'' nature of neural models sets a high barrier for externals to audit their training datasets, which further connives these unauthorized usages. Currently, watermarking methods have been proposed to prohibit inappropriate usage of image and natural language datasets. However, due to domain specificity, they are not directly applicable to code datasets, leaving the copyright protection of this emerging and important field of code data still exposed to threats. To fill this gap, we propose a method, named CodeMark, to embed user-defined imperceptible watermarks into code datasets to trace their usage in training neural code completion models. CodeMark is based on adaptive semantic-preserving transformations, which preserve the exact functionality of the code data and keep the changes covert against rule-breakers. We implement CodeMark in a toolkit and conduct an extensive evaluation of code completion models. CodeMark is validated to fulfill all desired properties of practical watermarks, including harmlessness to model accuracy, verifiability, robustness, and imperceptibility.

Guangke Chen, Yedi Zhang, Fu Song

Membership inference attacks allow adversaries to determine whether a particular example was contained in the model's training dataset. While previous works have confirmed the feasibility of such attacks in various applications, none has focused on speaker recognition (SR), a promising voice-based biometric recognition technique. In this work, we propose SLMIA-SR, the first membership inference attack tailored to SR. In contrast to conventional example-level attack, our attack features speaker-level membership inference, i.e., determining if any voices of a given speaker, either the same as or different from the given inference voices, have been involved in the training of a model. It is particularly useful and practical since the training and inference voices are usually distinct, and it is also meaningful considering the open-set nature of SR, namely, the recognition speakers were often not present in the training data. We utilize intra-similarity and inter-dissimilarity, two training objectives of SR, to characterize the differences between training and non-training speakers and quantify them with two groups of features driven by carefully-established feature engineering to mount the attack. To improve the generalizability of our attack, we propose a novel mixing ratio training strategy to train attack models. To enhance the attack performance, we introduce voice chunk splitting to cope with the limited number of inference voices and propose to train attack models dependent on the number of inference voices. Our attack is versatile and can work in both white-box and black-box scenarios. Additionally, we propose two novel techniques to reduce the number of black-box queries while maintaining the attack performance. Extensive experiments demonstrate the effectiveness of SLMIA-SR.

Deyuan Qu, Qi Chen, Yongqi Zhu et al.

In cooperative perception studies, there is often a trade-off between communication bandwidth and perception performance. While current feature fusion solutions are known for their excellent object detection performance, transmitting the entire sets of intermediate feature maps requires substantial bandwidth. Furthermore, these fusion approaches are typically limited to vehicles that use identical detection models. Our goal is to develop a solution that supports cooperative perception across vehicles equipped with different modalities of sensors. This method aims to deliver improved perception performance compared to late fusion techniques, while achieving precision similar to the state-of-art intermediate fusion, but requires an order of magnitude less bandwidth. We propose HEAD, a method that fuses features from the classification and regression heads in 3D object detection networks. Our method is compatible with heterogeneous detection networks such as LiDAR PointPillars, SECOND, VoxelNet, and camera Bird's-eye View (BEV) Encoder. Given the naturally smaller feature size in the detection heads, we design a self-attention mechanism to fuse the classification head and a complementary feature fusion layer to fuse the regression head. Our experiments, comprehensively evaluated on the V2V4Real and OPV2V datasets, demonstrate that HEAD is a fusion method that effectively balances communication bandwidth and perception performance.

Ye Tao, Wanwei Liu, Fu Song et al.

Deep neural networks, (DNNs, a.k.a. NNs), have been widely used in various tasks and have been proven to be successful. However, the accompanied expensive computing and storage costs make the deployments in resource-constrained devices a significant concern. To solve this issue, quantization has emerged as an effective way to reduce the costs of DNNs with little accuracy degradation by quantizing floating-point numbers to low-width fixed-point representations. Quantized neural networks (QNNs) have been developed, with binarized neural networks (BNNs) restricted to binary values as a special case. Another concern about neural networks is their vulnerability and lack of interpretability. Despite the active research on trustworthy of DNNs, few approaches have been proposed to QNNs. To this end, this paper presents an automata-theoretic approach to synthesizing BNNs that meet designated properties. More specifically, we define a temporal logic, called BLTL, as the specification language. We show that each BLTL formula can be transformed into an automaton on finite words. To deal with the state-explosion problem, we provide a tableau-based approach in real implementation. For the synthesis procedure, we utilize SMT solvers to detect the existence of a model (i.e., a BNN) in the construction process. Notably, synthesis provides a way to determine the hyper-parameters of the network before training.Moreover, we experimentally evaluate our approach and demonstrate its effectiveness in improving the individual fairness and local robustness of BNNs while maintaining accuracy to a great extent.

Mingchen Li, Heng Fan, Song Fu et al.

Prompt privacy is crucial, especially when using online large language models (LLMs), due to the sensitive information often contained within prompts. While LLMs can enhance prompt privacy through text rewriting, existing methods primarily focus on document-level rewriting, neglecting the rich, multi-granular representations of text. This limitation restricts LLM utilization to specific tasks, overlooking their generalization and in-context learning capabilities, thus hindering practical application. To address this gap, we introduce DP-GTR, a novel three-stage framework that leverages local differential privacy (DP) and the composition theorem via group text rewriting. DP-GTR is the first framework to integrate both document-level and word-level information while exploiting in-context learning to simultaneously improve privacy and utility, effectively bridging local and global DP mechanisms at the individual data point level. Experiments on CommonSense QA and DocVQA demonstrate that DP-GTR outperforms existing approaches, achieving a superior privacy-utility trade-off. Furthermore, our framework is compatible with existing rewriting techniques, serving as a plug-in to enhance privacy protection. Our code is publicly available at github.com/ResponsibleAILab/DP-GTR.

Yifan Jiao, Yunhao Li, Junhua Ding et al.

In this paper, we present a novel benchmark, GSOT3D, that aims at facilitating development of generic 3D single object tracking (SOT) in the wild. Specifically, GSOT3D offers 620 sequences with 123K frames, and covers a wide selection of 54 object categories. Each sequence is offered with multiple modalities, including the point cloud (PC), RGB image, and depth. This allows GSOT3D to support various 3D tracking tasks, such as single-modal 3D SOT on PC and multi-modal 3D SOT on RGB-PC or RGB-D, and thus greatly broadens research directions for 3D object tracking. To provide highquality per-frame 3D annotations, all sequences are labeled manually with multiple rounds of meticulous inspection and refinement. To our best knowledge, GSOT3D is the largest benchmark dedicated to various generic 3D object tracking tasks. To understand how existing 3D trackers perform and to provide comparisons for future research on GSOT3D, we assess eight representative point cloud-based tracking models. Our evaluation results exhibit that these models heavily degrade on GSOT3D, and more efforts are required for robust and generic 3D object tracking. Besides, to encourage future research, we present a simple yet effective generic 3D tracker, named PROT3D, that localizes the target object via a progressive spatial-temporal network and outperforms all current solutions by a large margin. By releasing GSOT3D, we expect to advance further 3D tracking in future research and applications. Our benchmark and model as well as the evaluation results will be publicly released at our webpage https://github.com/ailovejinx/GSOT3D.

Guangke Chen, Sen Chen, Lingling Fan et al.

Speaker recognition (SR) is widely used in our daily life as a biometric authentication or identification mechanism. The popularity of SR brings in serious security concerns, as demonstrated by recent adversarial attacks. However, the impacts of such threats in the practical black-box setting are still open, since current attacks consider the white-box setting only. In this paper, we conduct the first comprehensive and systematic study of the adversarial attacks on SR systems (SRSs) to understand their security weakness in the practical blackbox setting. For this purpose, we propose an adversarial attack, named FAKEBOB, to craft adversarial samples. Specifically, we formulate the adversarial sample generation as an optimization problem, incorporated with the confidence of adversarial samples and maximal distortion to balance between the strength and imperceptibility of adversarial voices. One key contribution is to propose a novel algorithm to estimate the score threshold, a feature in SRSs, and use it in the optimization problem to solve the optimization problem. We demonstrate that FAKEBOB achieves 99% targeted attack success rate on both open-source and commercial systems. We further demonstrate that FAKEBOB is also effective on both open-source and commercial systems when playing over the air in the physical world. Moreover, we have conducted a human study which reveals that it is hard for human to differentiate the speakers of the original and adversarial voices. Last but not least, we show that four promising defense methods for adversarial attack from the speech recognition domain become ineffective on SRSs against FAKEBOB, which calls for more effective defense methods. We highlight that our study peeks into the security implications of adversarial attacks on SRSs, and realistically fosters to improve the security robustness of SRSs.

Lei Bu, Yuchao Duan, Fu Song et al.

Numerous methods for crafting adversarial examples were proposed recently with high success rate. Since most existing machine learning based classifiers normalize images into some continuous, real vector, domain firstly, attacks often craft adversarial examples in such domain. However, "adversarial" examples may become benign after denormalizing them back into the discrete integer domain, known as the discretization problem. This problem was mentioned in some work, but has received relatively little attention. In this work, we first conduct a comprehensive study of existing methods and tools for crafting. We theoretically analyze 34 representative methods and empirically study 20 representative open source tools for crafting adversarial images. Our study reveals that the discretization problem is far more serious than originally thought. This suggests that the discretization problem should be taken into account seriously when crafting adversarial examples and measuring attack success rate. As a first step towards addressing this problem in black-box scenario, we propose a black-box method which reduces the adversarial example searching problem to a derivative-free optimization problem. Our method is able to craft adversarial images by derivative-free search in the discrete integer domain. Experimental results show that our method is comparable to recent white-box methods (e.g., FGSM, BIM and C\&W) and achieves significantly higher success rate in terms of adversarial examples in the discrete integer domain than recent black-box methods (e.g., ZOO, NES-PGD and Bandits). Moreover, our method is able to handle models that is non-differentiable and successfully break the winner of NIPS 2017 competition on defense with 95\% success rate. Our results suggest that discrete optimization algorithms open up a promising area of research into effective black-box attacks.

Yedi Zhang, Lei Huang, Pengfei Gao et al.

In the rapidly evolving landscape of neural network security, the resilience of neural networks against bit-flip attacks (i.e., an attacker maliciously flips an extremely small amount of bits within its parameter storage memory system to induce harmful behavior), has emerged as a relevant area of research. Existing studies suggest that quantization may serve as a viable defense against such attacks. Recognizing the documented susceptibility of real-valued neural networks to such attacks and the comparative robustness of quantized neural networks (QNNs), in this work, we introduce BFAVerifier, the first verification framework designed to formally verify the absence of bit-flip attacks or to identify all vulnerable parameters in a sound and rigorous manner. BFAVerifier comprises two integral components: an abstraction-based method and an MILP-based method. Specifically, we first conduct a reachability analysis with respect to symbolic parameters that represent the potential bit-flip attacks, based on a novel abstract domain with a sound guarantee. If the reachability analysis fails to prove the resilience of such attacks, then we encode this verification problem into an equivalent MILP problem which can be solved by off-the-shelf solvers. Therefore, BFAVerifier is sound, complete, and reasonably efficient. We conduct extensive experiments, which demonstrate its effectiveness and efficiency across various network architectures, quantization bit-widths, and adversary capabilities.

Zongxin Liu, Zhe Zhao, Fu Song et al.

Formal verification provides critical security assurances for neural networks, yet its practical application suffers from the long verification time. This work introduces a novel method for training verification-friendly neural networks, which are robust, easy to verify, and relatively accurate. Our method integrates neuron behavior consistency into the training process, making neuron activation states remain consistent across different inputs within a local neighborhood. This reduces the number of unstable neurons and tightens the bounds of neurons thereby enhancing the network's verifiability. We evaluated our method using the MNIST, Fashion-MNIST, and CIFAR-10 datasets with various network architectures. The experimental results demonstrate that networks trained using our method are verification-friendly across different radii and architectures, whereas other tools fail to maintain verifiability as the radius increases. Additionally, we show that our method can be combined with existing approaches to further improve the verifiability of networks.

Guangke Chen, Yedi Zhang, Zhe Zhao et al.

Current adversarial attacks against speaker recognition systems (SRSs) require either white-box access or heavy black-box queries to the target SRS, thus still falling behind practical attacks against proprietary commercial APIs and voice-controlled devices. To fill this gap, we propose QFA2SR, an effective and imperceptible query-free black-box attack, by leveraging the transferability of adversarial voices. To improve transferability, we present three novel methods, tailored loss functions, SRS ensemble, and time-freq corrosion. The first one tailors loss functions to different attack scenarios. The latter two augment surrogate SRSs in two different ways. SRS ensemble combines diverse surrogate SRSs with new strategies, amenable to the unique scoring characteristics of SRSs. Time-freq corrosion augments surrogate SRSs by incorporating well-designed time-/frequency-domain modification functions, which simulate and approximate the decision boundary of the target SRS and distortions introduced during over-the-air attacks. QFA2SR boosts the targeted transferability by 20.9%-70.7% on four popular commercial APIs (Microsoft Azure, iFlytek, Jingdong, and TalentedSoft), significantly outperforming existing attacks in query-free setting, with negligible effect on the imperceptibility. QFA2SR is also highly effective when launched over the air against three wide-spread voice assistants (Google Assistant, Apple Siri, and TMall Genie) with 60%, 46%, and 70% targeted transferability, respectively.

Guangke Chen, Zhe Zhao, Fu Song et al.

Adversarial attacks have been expanded to speaker recognition (SR). However, existing attacks are often assessed using different SR models, recognition tasks and datasets, and only few adversarial defenses borrowed from computer vision are considered. Yet,these defenses have not been thoroughly evaluated against adaptive attacks. Thus, there is still a lack of quantitative understanding about the strengths and limitations of adversarial attacks and defenses. More effective defenses are also required for securing SR systems. To bridge this gap, we present SEC4SR, the first platform enabling researchers to systematically and comprehensively evaluate adversarial attacks and defenses in SR. SEC4SR incorporates 4 white-box and 2 black-box attacks, 24 defenses including our novel feature-level transformations. It also contains techniques for mounting adaptive attacks. Using SEC4SR, we conduct thus far the largest-scale empirical study on adversarial attacks and defenses in SR, involving 23 defenses, 15 attacks and 4 attack settings. Our study provides lots of useful findings that may advance future research: such as (1) all the transformations slightly degrade accuracy on benign examples and their effectiveness vary with attacks; (2) most transformations become less effective under adaptive attacks, but some transformations become more effective; (3) few transformations combined with adversarial training yield stronger defenses over some but not all attacks, while our feature-level transformation combined with adversarial training yields the strongest defense over all the attacks. Extensive experiments demonstrate capabilities and advantages of SEC4SR which can benefit future research in SR.

Zhe Zhao, Guangke Chen, Jingyi Wang et al.

As a new programming paradigm, deep learning has expanded its application to many real-world problems. At the same time, deep learning based software are found to be vulnerable to adversarial attacks. Though various defense mechanisms have been proposed to improve robustness of deep learning software, many of them are ineffective against adaptive attacks. In this work, we propose a novel characterization to distinguish adversarial examples from benign ones based on the observation that adversarial examples are significantly less robust than benign ones. As existing robustness measurement does not scale to large networks, we propose a novel defense framework, named attack as defense (A2D), to detect adversarial examples by effectively evaluating an example's robustness. A2D uses the cost of attacking an input for robustness evaluation and identifies those less robust examples as adversarial since less robust examples are easier to attack. Extensive experiment results on MNIST, CIFAR10 and ImageNet show that A2D is more effective than recent promising approaches. We also evaluate our defence against potential adaptive attacks and show that A2D is effective in defending carefully designed adaptive attacks, e.g., the attack success rate drops to 0% on CIFAR10.

Yedi Zhang, Zhe Zhao, Guangke Chen et al.

Verifying and explaining the behavior of neural networks is becoming increasingly important, especially when they are deployed in safety-critical applications. In this paper, we study verification problems for Binarized Neural Networks (BNNs), the 1-bit quantization of general real-numbered neural networks. Our approach is to encode BNNs into Binary Decision Diagrams (BDDs), which is done by exploiting the internal structure of the BNNs. In particular, we translate the input-output relation of blocks in BNNs to cardinality constraints which are then encoded by BDDs. Based on the encoding, we develop a quantitative verification framework for BNNs where precise and comprehensive analysis of BNNs can be performed. We demonstrate the application of our framework by providing quantitative robustness analysis and interpretability for BNNs. We implement a prototype tool BDD4BNN and carry out extensive experiments which confirm the effectiveness and efficiency of our approach.

Wenjie Wan, Zhaodi Zhang, Yiwei Zhu et al.

Deep Neural Networks (DNNs) have become key components of many safety-critical applications such as autonomous driving and medical diagnosis. However, DNNs have been shown suffering from poor robustness because of their susceptibility to adversarial examples such that small perturbations to an input result in misprediction. Addressing to this concern, various approaches have been proposed to formally verify the robustness of DNNs. Most of these approaches reduce the verification problem to optimization problems of searching an adversarial example for a given input so that it is not correctly classified to the original label. However, they are limited in accuracy and scalability. In this paper, we propose a novel approach that can accelerate the robustness verification techniques by guiding the verification with target labels. The key insight of our approach is that the robustness verification problem of DNNs can be solved by verifying sub-problems of DNNs, one per target label. Fixing the target label during verification can drastically reduce the search space and thus improve the efficiency. We also propose an approach by leveraging symbolic interval propagation and linear relaxation techniques to sort the target labels in terms of chances that adversarial examples exist. This often allows us to quickly falsify the robustness of DNNs and the verification for remaining target labels could be avoided. Our approach is orthogonal to, and can be integrated with, many existing verification techniques. For evaluation purposes, we integrate it with three recent promising DNN verification tools, i.e., MipVerify, DeepZ, and Neurify. Experimental results show that our approach can significantly improve these tools by 36X speedup when the perturbation distance is set in a reasonable range.

Pengfei Gao, Hongyi Xie, Fu Song et al.

Side-channel attacks, which are capable of breaking secrecy via side-channel information, pose a growing threat to the implementation of cryptographic algorithms. Masking is an effective countermeasure against side-channel attacks by removing the statistical dependence between secrecy and power consumption via randomization. However, designing efficient and effective masked implementations turns out to be an error-prone task. Current techniques for verifying whether masked programs are secure are limited in their applicability and accuracy, especially when they are applied. To bridge this gap, in this article, we first propose a sound type system, equipped with an efficient type inference algorithm, for verifying masked arithmetic programs against higher-order attacks. We then give novel model-counting based and pattern-matching based methods which are able to precisely determine whether the potential leaky observable sets detected by the type system are genuine or simply spurious. We evaluate our approach on various implementations of arithmetic cryptographicprograms.The experiments confirm that our approach out performs the state-of-the-art base lines in terms of applicability, accuracy and efficiency.

Yusi Lei, Sen Chen, Lingling Fan et al.

Machine learning (ML) based approaches have been the mainstream solution for anti-phishing detection. When they are deployed on the client-side, ML-based classifiers are vulnerable to evasion attacks. However, such potential threats have received relatively little attention because existing attacks destruct the functionalities or appearance of webpages and are conducted in the white-box scenario, making it less practical. Consequently, it becomes imperative to understand whether it is possible to launch evasion attacks with limited knowledge of the classifier, while preserving the functionalities and appearance. In this work, we show that even in the grey-, and black-box scenarios, evasion attacks are not only effective on practical ML-based classifiers, but can also be efficiently launched without destructing the functionalities and appearance. For this purpose, we propose three mutation-based attacks, differing in the knowledge of the target classifier, addressing a key technical challenge: automatically crafting an adversarial sample from a known phishing website in a way that can mislead classifiers. To launch attacks in the white- and grey-box scenarios, we also propose a sample-based collision attack to gain the knowledge of the target classifier. We demonstrate the effectiveness and efficiency of our evasion attacks on the state-of-the-art, Google's phishing page filter, achieved 100% attack success rate in less than one second per website. Moreover, the transferability attack on BitDefender's industrial phishing page classifier, TrafficLight, achieved up to 81.25% attack success rate. We further propose a similarity-based method to mitigate such evasion attacks, Pelican. We demonstrate that Pelican can effectively detect evasion attacks. Our findings contribute to design more robust phishing website classifiers in practice.

Pengfei Gao, Fu Song, Taolue Chen et al.

JavaScript has become one of the most widely used languages for Web development. However, it is challenging to ensure the correctness and reliability of Web applications written in JavaScript, due to their dynamic and event-driven features. A variety of dynamic analysis techniques for JavaScript Web applications have been proposed, but they are limited in either coverage or scalability. In this paper, we propose a model-based automated approach to achieve high code coverage in a reasonable amount of time via testing with longer event sequences. We implement our approach as the tool LJS, and perform extensive experiments on 21 publicly available benchmarks (18,559 lines of code in total). On average, LJS achieves 86.4\% line coverage in 10 minutes, which is 5.4\% higher than that of JSDep, a breadth-first search based automated testing tool enriched with partial order reduction. In particular, on large applications, the coverage of LJS is 11-18\% higher than that of JSDep. Our empirical finding supports that longer test sequences can achieve higher code coverage in JavsScript testing.

Pengfei Gao, Hongyi Xie, Jun Zhang et al.

Power side-channel attacks, which can deduce secret data via statistical analysis, have become a serious threat. Masking is an effective countermeasure for reducing the statistical dependence between secret data and side-channel information. However, designing masking algorithms is an error-prone process. In this paper, we propose a hybrid approach combing type inference and model-counting to verify masked arithmetic programs against side-channel attacks. The type inference allows an efficient, lightweight procedure to determine most observable variables whereas model-counting accounts for completeness. In case that the program is not perfectly masked, we also provide a method to quantify the security level of the program. We implement our methods in a tool QMVerif and evaluate it on cryptographic benchmarks. The experimental results show the effectiveness and efficiency of our approach.

Yedi Zhang, Fu Song, Taolue Chen

Alternating-time temporal logics (ATL/ATL*) represent a family of modal logics for reasoning about agents' strategic abilities in multiagent systems (MAS). The interpretations of ATL/ATL* over the semantic model Concurrent Game Structures (CGS) usually vary depending on the agents' abilities, for instance, perfect vs. imperfect information, perfect vs. imperfect recall, resulting in a variety of variants which have been studied extensively in literature. However, they are defined at the semantic level, which may limit modeling flexibilities and may give counter-intuitive interpretations. To mitigate these issues, in this work, we propose to extend CGS with agents' abilities and study the new semantics of ATL/ATL* under this model. We give PSACE/2EXPTIME model-checking algorithms for ATL/ATL* and implement them as a prototype tool. Experiment results show the practical feasibility of the approach.

Fu Song, Tayssir Touili

A Dynamic Pushdown Network (DPN) is a set of pushdown systems (PDSs) where each process can dynamically create new instances of PDSs. DPNs are a natural model of multi-threaded programs with (possibly recursive) procedure calls and thread creation. Extending DPNs with locks allows processes to synchronize with each other. Thus, DPNs with locks are a well adapted formalism to model multi-threaded programs that synchronize via locks. Therefore, it is important to have model-checking algorithms for DPNs with locks. We consider in this work model-checking for DPNs with locks against single-indexed LTL properties of the form V fi s.t. fi is a LTL formula interpreted over the PDS i. We consider the model-checking problems w.r.t. simple valuations (i.e, whether a configuration satisfies an atomic proposition depends only on its control location and held locks) and w.r.t. regular valuations (i.e., the set of the configurations satisfying an atomic proposition is a regular set of configurations). We show that these model-checking problems are decidable.