Search

With the development of deep learning processors and accelerators, deep learning models have been widely deployed on edge devices as part of the Internet of Things. Edge device models are generally considered as valuable intellectual properties that are worth for careful protection. Unfortunately, these models have a great risk of being stolen or illegally copied. The existing model protections using encryption algorithms are suffered from high computation overhead which is not practical due to the limited computing capacity on edge devices. In this work, we propose a light-weight, practical, and general Edge device model Pro tection method at neuron level, denoted as EdgePro. Specifically, we select several neurons as authorization neurons and set their activation values to locking values and scale the neuron outputs as the "asswords" during training. EdgePro protects the model by ensuring it can only work correctly when the "passwords" are met, at the cost of encrypting and storing the information of the "passwords" instead of the whole model. Extensive experimental results indicate that EdgePro can work well on the task of protecting on datasets with different modes. The inference time increase of EdgePro is only 60% of state-of-the-art methods, and the accuracy loss is less than 1%. Additionally, EdgePro is robust against adaptive attacks including fine-tuning and pruning, which makes it more practical in real-world applications. EdgePro is also open sourced to facilitate future research: https://github.com/Leon022/Edg

Trojan attack on deep neural networks, also known as backdoor attack, is a typical threat to artificial intelligence. A trojaned neural network behaves normally with clean inputs. However, if the input contains a particular trigger, the trojaned model will have attacker-chosen abnormal behavior. Although many backdoor detection methods exist, most of them assume that the defender has access to a set of clean validation samples or samples with the trigger, which may not hold in some crucial real-world cases, e.g., the case where the defender is the maintainer of model-sharing platforms. Thus, in this paper, we propose FreeEagle, the first data-free backdoor detection method that can effectively detect complex backdoor attacks on deep neural networks, without relying on the access to any clean samples or samples with the trigger. The evaluation results on diverse datasets and model architectures show that FreeEagle is effective against various complex backdoor attacks, even outperforming some state-of-the-art non-data-free backdoor detection methods., Comment: Accepted by USENIX Security 2023

In recent years, REST API fuzzing has emerged to explore errors on a cloud service. Its performance highly depends on the sequence construction and request generation. However, existing REST API fuzzers have trouble generating long sequences with well-constructed requests to trigger hard-to-reach states in a cloud service, which limits their performance of finding deep errors and security bugs. Further, they cannot find the specific errors caused by using undefined parameters during request generation. Therefore, in this paper, we propose a novel hybrid data-driven solution, named MINER, with three new designs working together to address the above limitations. First, MINER collects the valid sequences whose requests pass the cloud service's checking as the templates, and assigns more executions to long sequence templates. Second, to improve the generation quality of requests in a sequence template, MINER creatively leverages the state-of-the-art neural network model to predict key request parameters and provide them with appropriate parameter values. Third, MINER implements a new data-driven security rule checker to capture the new kind of errors caused by undefined parameters. We evaluate MINER against the state-of-the-art fuzzer RESTler on GitLab, Bugzilla, and WordPress via 11 REST APIs. The results demonstrate that the average pass rate of MINER is 23.42% higher than RESTler. MINER finds 97.54% more unique errors than RESTler on average and 142.86% more reproducible errors after manual analysis. We have reported all the newly found errors, and 7 of them have been confirmed as logic bugs by the corresponding vendors., Comment: Accepted as a full paper at USENIX Security '23

Currently, natural language processing (NLP) models are wildly used in various scenarios. However, NLP models, like all deep models, are vulnerable to adversarially generated text. Numerous works have been working on mitigating the vulnerability from adversarial attacks. Nevertheless, there is no comprehensive defense in existing works where each work targets a specific attack category or suffers from the limitation of computation overhead, irresistible to adaptive attack, etc. In this paper, we exhaustively investigate the adversarial attack algorithms in NLP, and our empirical studies have discovered that the attack algorithms mainly disrupt the importance distribution of words in a text. A well-trained model can distinguish subtle importance distribution differences between clean and adversarial texts. Based on this intuition, we propose TextDefense, a new adversarial example detection framework that utilizes the target model's capability to defend against adversarial attacks while requiring no prior knowledge. TextDefense differs from previous approaches, where it utilizes the target model for detection and thus is attack type agnostic. Our extensive experiments show that TextDefense can be applied to different architectures, datasets, and attack methods and outperforms existing methods. We also discover that the leading factor influencing the performance of TextDefense is the target model's generalizability. By analyzing the property of the target model and the property of the adversarial example, we provide our insights into the adversarial attacks in NLP and the principles of our defense method.

Nowadays, IoT devices integrate a wealth of third-party components (TPCs) in firmware to shorten the development cycle. TPCs usually have strict usage specifications, e.g., checking the return value of the function. Violating the usage specifications of TPCs can cause serious consequences, e.g., NULL pointer dereference. Therefore, this massive amount of TPC integrations, if not properly implemented, will lead to pervasive vulnerabilities in IoT devices. Detecting vulnerabilities automatically in TPC integration is challenging from several perspectives: (1) There is a gap between the high-level specifications from TPC documents, and the low-level implementations in the IoT firmware. (2) IoT firmware is mostly the closed-source binary, which loses a lot of information when compiling from the source code and has diverse architectures. To address these challenges, we design and implement UVScan, an automated and scalable system to detect TPC usage violations in IoT firmware. In UVScan, we first propose a novel natural language processing (NLP)-based rule extraction framework, which extracts API specifications from inconsistently formatted TPC documents. We then design a rule-driven NLP-guided binary analysis engine, which maps the logical information from the high-level TPC document to the low-level binary, and detects TPC usage violations in IoT firmware across different architectures. We evaluate UVScan from four perspectives on four popular TPCs and six ground-truth datasets. The results show that UVScan achieves more than 70% precision and recall, and has a significant performance improvement compared with even the source-level API misuse detectors., Comment: Accepted as a full paper at USENIX Security '23

Recently, face swapping has been developing rapidly and achieved a surprising reality, raising concerns about fake content. As a countermeasure, various detection approaches have been proposed and achieved promising performance. However, most existing detectors struggle to maintain performance on unseen face swapping methods and low-quality images. Apart from the generalization problem, current detection approaches have been shown vulnerable to evasion attacks crafted by detection-aware manipulators. Lack of robustness under adversary scenarios leaves threats for applying face swapping detection in real world. In this paper, we propose a novel face swapping detection approach based on face identification probability distributions, coined as IdP_FSD, to improve the generalization and robustness. IdP_FSD is specially designed for detecting swapped faces whose identities belong to a finite set, which is meaningful in real-world applications. Compared with previous general detection methods, we make use of the available real faces with concerned identities and require no fake samples for training. IdP_FSD exploits face swapping's common nature that the identity of swapped face combines that of two faces involved in swapping. We reflect this nature with the confusion of a face identification model and measure the confusion with the maximum value of the output probability distribution. What's more, to defend our detector under adversary scenarios, an attention-based finetuning scheme is proposed for the face identification models used in IdP_FSD. Extensive experiments show that the proposed IdP_FSD not only achieves high detection performance on different benchmark datasets and image qualities but also raises the bar for manipulators to evade the detection.

Vertical federated learning is a trending solution for multi-party collaboration in training machine learning models. Industrial frameworks adopt secure multi-party computation methods such as homomorphic encryption to guarantee data security and privacy. However, a line of work has revealed that there are still leakage risks in VFL. The leakage is caused by the correlation between the intermediate representations and the raw data. Due to the powerful approximation ability of deep neural networks, an adversary can capture the correlation precisely and reconstruct the data. To deal with the threat of the data reconstruction attack, we propose a hashing-based VFL framework, called \textit{HashVFL}, to cut off the reversibility directly. The one-way nature of hashing allows our framework to block all attempts to recover data from hash codes. However, integrating hashing also brings some challenges, e.g., the loss of information. This paper proposes and addresses three challenges to integrating hashing: learnability, bit balance, and consistency. Experimental results demonstrate \textit{HashVFL}'s efficiency in keeping the main task's performance and defending against data reconstruction attacks. Furthermore, we also analyze its potential value in detecting abnormal inputs. In addition, we conduct extensive experiments to prove \textit{HashVFL}'s generalization in various settings. In summary, \textit{HashVFL} provides a new perspective on protecting multi-party's data security and privacy in VFL. We hope our study can attract more researchers to expand the application domains of \textit{HashVFL}.

Deep neural networks (DNNs) have demonstrated their outperformance in various domains. However, it raises a social concern whether DNNs can produce reliable and fair decisions especially when they are applied to sensitive domains involving valuable resource allocation, such as education, loan, and employment. It is crucial to conduct fairness testing before DNNs are reliably deployed to such sensitive domains, i.e., generating as many instances as possible to uncover fairness violations. However, the existing testing methods are still limited from three aspects: interpretability, performance, and generalizability. To overcome the challenges, we propose NeuronFair, a new DNN fairness testing framework that differs from previous work in several key aspects: (1) interpretable - it quantitatively interprets DNNs' fairness violations for the biased decision; (2) effective - it uses the interpretation results to guide the generation of more diverse instances in less time; (3) generic - it can handle both structured and unstructured data. Extensive evaluations across 7 datasets and the corresponding DNNs demonstrate NeuronFair's superior performance. For instance, on structured datasets, it generates much more instances (~x5.84) and saves more time (with an average speedup of 534.56%) compared with the state-of-the-art methods. Besides, the instances of NeuronFair can also be leveraged to improve the fairness of the biased DNNs, which helps build more fair and trustworthy deep learning systems.

Towards real-world information extraction scenario, research of relation extraction is advancing to document-level relation extraction(DocRE). Existing approaches for DocRE aim to extract relation by encoding various information sources in the long context by novel model architectures. However, the inherent long-tailed distribution problem of DocRE is overlooked by prior work. We argue that mitigating the long-tailed distribution problem is crucial for DocRE in the real-world scenario. Motivated by the long-tailed distribution problem, we propose an Easy Relation Augmentation(ERA) method for improving DocRE by enhancing the performance of tailed relations. In addition, we further propose a novel contrastive learning framework based on our ERA, i.e., ERACL, which can further improve the model performance on tailed relations and achieve competitive overall DocRE performance compared to the state-of-arts.

Additional file 1.

Vertical federated learning (VFL) is an emerging paradigm that enables collaborators to build machine learning models together in a distributed fashion. In general, these parties have a group of users in common but own different features. Existing VFL frameworks use cryptographic techniques to provide data privacy and security guarantees, leading to a line of works studying computing efficiency and fast implementation. However, the security of VFL's model remains underexplored.

Additional file 2 .

Facial Liveness Verification (FLV) is widely used for identity authentication in many security-sensitive domains and offered as Platform-as-a-Service (PaaS) by leading cloud vendors. Yet, with the rapid advances in synthetic media techniques (e.g., deepfake), the security of FLV is facing unprecedented challenges, about which little is known thus far. To bridge this gap, in this paper, we conduct the first systematic study on the security of FLV in real-world settings. Specifically, we present LiveBugger, a new deepfake-powered attack framework that enables customizable, automated security evaluation of FLV. Leveraging LiveBugger, we perform a comprehensive empirical assessment of representative FLV platforms, leading to a set of interesting findings. For instance, most FLV APIs do not use anti-deepfake detection; even for those with such defenses, their effectiveness is concerning (e.g., it may detect high-quality synthesized videos but fail to detect low-quality ones). We then conduct an in-depth analysis of the factors impacting the attack performance of LiveBugger: a) the bias (e.g., gender or race) in FLV can be exploited to select victims; b) adversarial training makes deepfake more effective to bypass FLV; c) the input quality has a varying influence on different deepfake techniques to bypass FLV. Based on these findings, we propose a customized, two-stage approach that can boost the attack success rate by up to 70%. Further, we run proof-of-concept attacks on several representative applications of FLV (i.e., the clients of FLV APIs) to illustrate the practical implications: due to the vulnerability of the APIs, many downstream applications are vulnerable to deepfake. Finally, we discuss potential countermeasures to improve the security of FLV. Our findings have been confirmed by the corresponding vendors., Comment: Accepted as a full paper at USENIX Security '22

Additional file 1 .

Ternary nickel-cobalt lithium aluminate LiNixCoyAl1‐x‐yO2 (NCA, x≥0.8) is an essential cathode material with many vital advantages, such as lower cost and higher specific capacity compared with lithium cobaltate and lithium iron phosphate materials. However, the noticeably irreversible capacity and reduced cycle performance of NCA cathode materials have restricted their further development. To solve these problems and further improve the electrochemical performance, numerous research studies on material modification have been conducted, achieving promising results in recent years. In this work, the progress of NCA cathode materials is examined from the aspects of surface coating and bulk doping. Furthermore, future research directions for NCA cathode materials are proposed.

The pervasive usage of mobile phones provides an abundance of data associated with their users. Data generated by mobile devices includes measurements of user actions (e.g. call logs data, SMS logs data, app usage sessions data) as well as records regarding its local environment (e.g. data about other local blue tooth devices, GPS/location trajectories). These data provide information regarding the underlying behaviors of device users and their corresponding social contexts, thus giving us new opportunities for learning human spatio-temporal and social patterns.In this dissertation, new methods are developed to mine data generated by mobile devices and further improve applications in the area of social learning and inference. Particularly, three objectives are pursued here: (1) using mobile phone data to distinguish dyads embedded in social relationships from those that are not; (2) understanding how human telecommunication activities are associated with urban ecology; and (3) obtaining insights regarding users' demographic features from mobile app usage data. A key characteristic of these approaches is that they leverage activity and environmental data at the device level (in some cases, aggregated across sets of users), often allowing the data in question to be collected and stored in an efficient and privacy preserving manner. In pursuing the first objective, we introduce a new approach --\emph{activity correlation spectroscopy} -- to inferring relationships by exploiting the spectral and distributional structure of activity correlation within dyads. Unlike existing techniques, our approach can be employed with minimal, individual-level (i.e., non-relational), and non-identifying data that is easily collected using commodity hardware. We demonstrate our methodology via an application to detection of friendship and group co-membership using mobile device and survey data from the MIT Reality Mining study \citep{eagle2009}.Vis a vis the second objective, we provide a novel approach to the use of spatio-temporally aggregated cell phone data to learn features of urban ecology (i.e., spatial distributions of distinct social and economic entities and their associated activities). Specifically, our technique involves four stages: (\RN{1}) decomposing the aggregated cell phone activity within local areal units using spectral methods; (\RN{2}) learning spectral characteristics associated with ecological features using a training set; (\RN{3}) predicting local ecology composition for out-of-sample areas; and (\RN{4}) predicting activity time series for out-of-sample areas. The core of our approach is the projection of spectral features in cell phone activity series into an ecology-associated basis, allowing both identification of communication patterns arising from particular types of local activities and/or institutions and leveraging of those patterns for classification and activity prediction. We apply our methodology to aggregated communication and Internet traffic data from the cities of Milan and Trento to show the effectiveness of our method. Finally, in pursuing the third objective, we demonstrate an integrated system which can cheaply and easily collect application behavior and survey data from mobile phones; we introduce several novel features that assist the learning of individual level demographic features (e.g., gender and age group). Specifically, our approach for learning and inference for demographic features involves new techniques: (\RN{1}) decomposing the app usage from mobile phones using spectral methods; (\RN{2}) learning spectral characteristics associated with individuals using a training set; (\RN{3}) combining other temporal features with learned spectral characteristics to predict demographic features for out-of-sample individuals. The core of our methodology is the utilization of spectral features in cell phone app activity series, allowing both identification of behavior patterns arising from particular types of cell phone apps and leveraging of those patterns for demographic classification and prediction. We demonstrate the effectiveness of our approach with an application to real mobile app traffic data from the United States.

Redox status affects various cellular activities, such as proliferation, differentiation, and death. Recent studies suggest pivotal roles of reactive oxygen species not only in pathogenesis under oxidative insult but also in intracellular signal transduction. Glutathione is present in several millimolar concentrations in the cytoplasm and has multiple roles in the regulation of cellular homeostasis. Two enzymes, γ-glutamylcysteine synthetase and glutathione synthetase, constitute the de novo synthesis machinery, while glutathione reductase is involved in the recycling of oxidized glutathione. Multidrug resistant proteins and some other transporters are responsible for exporting oxidized glutathione, glutathione conjugates, and S-nitrosoglutathione. In addition to antioxidation, glutathione is more positively involved in cellular activity via its sulfhydryl moiety of a molecule. Animals in which genes responsible for glutathione metabolism are genetically modified can be used as beneficial and reliable models to elucidate roles of glutathione in vivo. This review article overviews recent progress in works related to genetically modified rodents and advances in the elucidation of glutathione-mediated reactions.

Teaching venues are playing an important role in the filed of open and distance education (ODL). In China's ODL system over 70% enrolled students are from Radio and TV Universities (RTVUs). Many off-campus teaching venues of e-colleges in 67 conventional universities have been established in RTVUs at prefecture and city level. In this article, the authors have done a direct survey on the data of distance education cost over the past consecutive 4 years in 21 RTVUs at prefecture and city level from 3 provinces (Xinjiang, Hunan and Fujian) in China's western, central and eastern areas, and empirically studied the scale effect in teaching venues at prefecture and city level under the hypothesis condition-teaching effect of ODL and conventional education is the same. The study result illustrates that the average student cost of RTVUs at prefecture and city level demonstrates a scale economy feature, that is, with the expansion of enrolled students scale, there is a fall in the average student cost. The study indicates large differences between average student variable cost and average student cost, which shows a possible average cost reduction in RTVUs at prefecture and city level. In terms of geographical and economic differences, some imbalances of development have been found among RTVUs at prefecture and city level. A quadratic form regression model has been established for doing this research in the perspective of education economics. The results of the regression show that the optimal enrolled student number of RTVUs nationwide at prefecture and city level is around 7,800. For provinces, the most optimal enrolled student numbers are as follows: 9,100 for Fujian, 3,800 for Hunan, and 2,200for Xianjiang, which can serve as the reference for expanding scale economy in ODL. The research finding also indicate the changes in student average cost resulting from the changes in scale, the problem of determining investment direction. It is suggested that more input which benefit the quality improvement of distance education should be made.

One of the main applications of wireless sensor networks is to provide proper coverage of their deployment regions. In this paper, we discussed the tolerance coverage area based node scheduling algorithm, and then improved this algorithm by two ways: improving the coverage redundancy judgment strategy by introducing the relative remaining energy level(RREL), and optimizing node state distribution by increasing a node states "pre-active", and then proposed a node state optimization based coverage control algorithm NSO-CC. Simulation results show that NSO-CC effectively solves the problem of nodes non-uniform sleeping, improves the network coverage quality and prolongs the network lifetime.

Nasopharyngeal carcinoma (NPC) is endemic in Southeast Asia. Although dendritic cell (DC)-based vaccine has emerged as a promising immunotherapy for various malignancies, its use in pediatric nasopharyngeal carcinoma (PNPC) has not been addressed. In this study, DCs isolated from peripheral blood monocytes of three pediatric patients with advanced (stage IV) NPC were incubated with whole-tumor-antigen preparations and differentiated into immature DCs in the presence of granulocyte-macrophage colony-stimulating factor and interleukin-4, and then underwent maturation when exposed to tumor necrosis factor-alpha. Upon maturation, DCs acquired the ability to stimulate T-cell proliferation as examined by [(3)H]-thymidine incorporation and the ability of these T-cells to secrete interferon-gamma as determined by enzyme-linked immunosorbent spot assay. Cytotoxic assay revealed that mature tumor-antigen-pulsed DCs induced cytotoxic activity of the T-cells against both autologous and allogeneic NPC tumor cells (including NPC tumor cells from a different individual or from CNE-2Z, a poorly differentiated human NPC cell line). Blocking HLA class I molecules by W6/32 inhibited T-cell-mediated cytotoxic activity in both autologous and allogeneic settings. Our results indicate that DCs pulsed with whole-tumor-antigen can effectively activate HLA class I-restricted cytotoxic T-cells in vitro, and thus provide experimental basis for their future clinical use in PNPC.

The saturated iron core superconducting fault current limiter (SCFCL) utilizes the non-linear of ferromagnetic hysteresis to change its inductance. Its cores are saturated deeply under normal operation but driven out of saturation whenever fault occurs, so it is very difficult to design this SCFCL through theoretical calculation and there will be biggish error if deal the magnetization curve with subsection linearization. In this paper, incorporating the Jiles-Atherton hysteresis model which can describe the actual measured hysteresis very well and constructing the dynamic differential equations of a circuit comprising this SCFCL, we put forward a new design method by numerical calculating the dynamic equations with computer. We also set forth the systematization design step and introduce a model designed using the technique. The design computation curves, which are agree well with the experimental date, are presented too. This paper shows that the design method is effective and can well forecast the limiting-current result of this SCFCL

The 3-phase reactor superconducting fault current limiter (SCFCL) consists of three superconducting windings with the same number of turns, which are wound on a single iron core. In a normal condition, it is a good conductor with no impedance because the flux is zero in the iron core. However, under a single-line-to-ground fault, it becomes a reactor with large inductance suppressing the fault current, Furthermore, for two or three-phase short circuits, the reactance of the SCFCL keep zero but it is good for limiting the fault current by the abruptly increasing resistance of the superconducting winding. In this paper, its equivalent impedance under every operating condition has been concluded and the simulations of short-circuit current limiting performance of designed prototype were carried out. The results are in good agreement with the theoretical analysis.

Actual power quality signal which is often affected by noise pollution impacts the analysis results of the disturbance signal. In this paper, EEMD (Ensemble Empirical Mode Decomposition) -based threshold de-noising method is proposed for power quality signal with different SNR ( S ignal-to- N oise R atio ). A s a comparison, we use other four thresholds, namely, the heuristic threshold, the self-adaptive threshold, the fixed threshold and the minimax threshold to filter the noises from power quality signal. Through the analysis and comparison of three characteristics of the signal pre-and-post de-noised, including waveforms, SNR and MSE ( M ean S quare E rror), furthermore the instantaneous attribute of corresponding time by HHT (Hilbert Huang T ransform). Simulation results show that EEMD threshold de-noising method can make the wavefor m clos e to the actual value . T he SNR is high er and the MSE is small er c ompared wit h other four thresholds . T he instantaneous attribute can reflect the actual disturbance signal more exactly. The o ptimal threshold EEMD -based algorithm is proposed for power quality disturbance signal de-noising. Meanwhile, EEMD threshold de-noising method with adaptivity is suitable for composite disturbance signal de-noising.

Facilitated by messaging protocols (MP), many home devices are connected to the Internet, bringing convenience and accessibility to customers. However, most deployed MPs on IoT platforms are fragmented and are not implemented carefully to support secure communication. To the best of our knowledge, there is no systematic solution to perform automatic security checks on MP implementations yet. To bridge the gap, we present MPInspector, the first automatic and systematic solution for vetting the security of MP implementations. MPInspector combines model learning with formal analysis and operates in three stages: (a) using parameter semantics extraction and interaction logic extraction to automatically infer the state machine of an MP implementation, (b) generating security properties based on meta properties and the state machine, and (c) applying automatic property based formal verification to identify property violations. We evaluate MPInspector on three popular MPs, including MQTT, CoAP and AMQP, implemented on nine leading IoT platforms. It identifies 252 property violations, leveraging which we further identify eleven types of attacks under two realistic attack scenarios. In addition, we demonstrate that MPInspector is lightweight (the average overhead of end-to-end analysis is ~4.5 hours) and effective with a precision of 100% in identifying property violations., Comment: Accepted as a full paper at USENIX Security '21