Deep Learning-based Text Understanding (DLTU) is the backbone technique behind various applications, including question answering, machine translation, and text classification.Despite its tremendous popularity, security vulnerabilities of DLTU are still largely unknown, which highly concerning given increasing use in security-sensitive applications such as sentiment analysis toxic content detection.In this paper, we show that inherently vulnerable to adversarial attacks, maliciously crafted...

This paper attacks the challenging problem of zero-example video retrieval. In such a retrieval paradigm, an end user searches for unlabeled videos by ad-hoc queries described in natural language text with no visual example provided. Given as sequences frames and words, effective sequence-to-sequence cross-modal matching is required. The majority existing methods are concept based, extracting relevant concepts from accordingly establishing associations between two modalities. contrast, this...

Traffic flow prediction plays an important role in ITS (Intelligent Transportation System). This task is challenging due to the complex spatial and temporal correlations (e.g., constraints of road network law dynamic change with time). Existing work tried solve this problem by exploiting a variety spatiotemporal models. However, we observe that more semantic pair-wise among possibly distant roads are also critical for traffic prediction. To jointly model spatial, temporal, various global...

Many of today's machine learning (ML) systems are built by reusing an array of, often pre-trained, primitive models, each fulfilling distinct functionality (e.g., feature extraction). The increasing use models significantly simplifies and expedites the development cycles ML systems. Yet, because most such contributed maintained untrusted sources, their lack standardization or regulation entails profound security implications, about which little is known thus far. In this paper, we...

Multi-frame human pose estimation in complicated situations is challenging. Although state-of-the-art joints detectors have demonstrated remarkable results for static images, their performances come short when we apply these models to video sequences. Prevalent shortcomings include the failure handle motion blur, defocus, or occlusions, arising from inability capturing temporal dependency among frames. On other hand, directly employing conventional recurrent neural networks incurs empirical...

Deep learning (DL) models are inherently vulnerable to adversarial examples - maliciously crafted inputs trigger target DL misbehave which significantly hinders the application of in security-sensitive domains. Intensive research on has led an arms race between adversaries and defenders. Such plethora emerging attacks defenses raise many questions: Which more evasive, preprocessing-proof, or transferable? effective, utility-preserving, general? Are ensembles multiple robust than individuals?...

In recent years, real-world attacks against PKI take place frequently. For example, malicious domains' certificates issued by compromised CAs are widespread, and revoked still trusted clients. spite of a lot research to improve the security SSL/TLS connections, there some problems unsolved. On one hand, although log-based schemes provided certificate audit service quickly detect CAs' misbehavior, data consistency log servers ignored. other checking is neglected due incomplete, insecure...

Nowadays, many computer and communication systems generate graph data. Graph data span different domains, ranging from online social network networks like Facebook to epidemiological used study the spread of infectious diseases. are shared regularly for purposes including academic research business collaborations. Since may be sensitive, owners often use various anonymization techniques that compromise resulting utility anonymized To make matters worse, there several state-of-the-art...

Recently, a new paradigm of building general-purpose language models (e.g., Google's Bert and OpenAI's GPT-2) in Natural Language Processing (NLP) for text feature extraction, standard procedure NLP systems that converts texts to vectors (i.e., embeddings) downstream modeling, has arisen starts find its application various tasks real world search engine [6]). To obtain embeddings, these have highly complicated architectures with millions learnable parameters are usually pretrained on...

In this paper, we study the quantification, practice, and implications of structural data (e.g., social data, mobility traces) De-Anonymization (DA). First, address several open problems in DA by quantifying perfect (1-ε)-perfect DA}, where ε is error tolerated a scheme. To best our knowledge, first work on under general model, which closes gap between practice theory. Second, conduct large-scale de-anonymizability 26 real world datasets, including Social Networks (SNs), Collaborations...

Despite their immense popularity, deep learning-based acoustic systems are inherently vulnerable to adversarial attacks, wherein maliciously crafted audios trigger target misbehave. In this paper, we present SirenAttack, a new class of attacks generate audios. Compared with existing SirenAttack highlights set significant features: (i) versatile -- it is able deceive range end-to-end under both white-box and black-box settings; (ii) effective that can be recognized as specific phrases by...

With the explosive development of information technology, vulnerabilities have become one major threats to computer security. Most with similar patterns can be detected effectively by static analysis methods. However, some vulnerable and non-vulnerable code is hardly distinguishable, resulting in low detection accuracy. In this paper, we define accurate identification as a fine-grained vulnerability problem. We propose VulSniper which designed detect more effectively. VulSniper, attention...

Recent years have witnessed the emergence of a new paradigm building natural language processing (NLP) systems: general-purpose, pre-trained models (LMs) are composed with simple downstream and fine-tuned for variety NLP tasks. This shift significantly simplifies system development cycles. However, as many LMs provided by untrusted third parties, their lack standardization or regulation entails profound security implications, which largely unexplored. To bridge this gap, work studies threats...

Despite their tremendous success in a range of domains, deep learning systems are inherently susceptible to two types manipulations: adversarial inputs -- maliciously crafted samples that deceive target neural network (DNN) models, and poisoned models adversely forged DNNs misbehave on pre-defined inputs. While prior work has intensively studied the attack vectors parallel, there is still lack understanding about fundamental connections: what dynamic interactions between vectors?...

Smart contracts hold digital coins worth billions of dollars, their security issues have drawn extensive attention in the past years. Towards smart contract vulnerability detection, conventional methods heavily rely on fixed expert rules, leading to low accuracy and poor scalability. Recent deep learning approaches alleviate this issue but fail encode useful knowledge. In paper, we explore combining with patterns an explainable fashion. Specifically, develop automatic tools extract from...

Abstract The low viability during gastrointestinal transit and poor mucoadhesion considerably limits the effectiveness of Ligilactobacillus salivarius Li01 (Li01) in regulating gut microbiota alleviating inflammatory bowel disease (IBD). In this study, a delivery system was designed through layer-by-layer (LbL) encapsulating single Li01cell with chitosan alginate. layers were strengthened by cross-linking to form firm mucoadhesive shell (~10 nm thickness) covering bacterial cell. LbL...

Predicting human motion from historical pose sequence is crucial for a machine to succeed in intelligent interactions with humans. One aspect that has been obviated so far, the fact how we represent skeletal critical impact on prediction results. Yet there no effort investigates across different representation schemes. We conduct an indepth study various representations focus their effects task. Moreover, recent approaches build upon off-the-shelf RNN units prediction. These process input...

Federated learning (FL) has emerged as a privacy-aware collaborative paradigm where participants jointly train powerful model without sharing their private data. One desirable property for FL is the implementation of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">right to be forgotten (RTBF)</i> , i.e., leaving participant right request deletion its data from global model. However, xmlns:xlink="http://www.w3.org/1999/xlink">unlearning itself...

As the first defensive layer that attacks would hit, web application firewall (WAF) plays an indispensable role in defending against malicious like SQL injection (SQLi). With development of cloud computing, WAF-as-a-service, as one kind Security-as-a-service, has been proposed to facilitate deployment, configuration, and update WAFs cloud. Despite its tremendous popularity, security vulnerabilities WAF-as-a-service are still largely unknown, which is highly concerning given massive usage. In...

Providing explanations for deep neural network (DNN) models is crucial their use in security-sensitive domains. A plethora of interpretation have been proposed to help users understand the inner workings DNNs: how does a DNN arrive at specific decision given input? The improved interpretability believed offer sense security by involving human decision-making process. Yet, due its data-driven nature, itself potentially susceptible malicious manipulations, about which little known thus far....

In this paper, we consider the problem of multiparty deep learning (MDL), wherein autonomous data owners jointly train accurate neural network models without sharing their private data. We design, implement, and evaluate ∝MDL, a new MDL paradigm built upon three primitives: asynchronous optimization, lightweight homomorphic encryption, threshold secret sharing. Compared with prior work, ∝MDL departs in significant ways: a) besides providing explicit privacy guarantee, it retains desirable...