SCIE, Ei, INSPEC, JST, AJ, MR, CA, DBLP, etc.
Edited by: Editorial Board of Journal Of Computer Science and Technology
P.O. Box 2704, Beijing 100190, P.R. China Sponsored by: Institute of Computing Technology, CAS & China Computer Federation Undertaken by: Institute of Computing Technology, CAS Published by: SCIENCE PRESS, BEIJING, CHINA Distributed by: China: All Local Post Offices Other Countries: Springer
Although using convolutional neural networks (CNN) for computer-aided diagnosis (CAD) has made tremendous progress in the last few years, the small medical datasets remain to be the major bottleneck in this area. To address this problem, researchers start looking for information out of the medical datasets. Previous efforts mainly leverage information from natural images via transfer learning. More recent research work focuses on integrating knowledge from medical practitioners, either letting networks resemble how practitioners are trained, how they view images, or using extra annotations. In this paper, we propose a scheme named Domain Guided-CNN (DG-CNN) to incorporate the margin information, a feature described in the consensus for radiologists to diagnose cancer in breast ultrasound (BUS) images. In DG-CNN, attention maps that highlight margin areas of tumors are first generated, and then incorporated via different approaches into the networks. We have tested the performance of DG-CNN on our own dataset (including 1485 ultrasound images) and on a public dataset. The results show that DG-CNN can be applied to different network structures like VGG and ResNet to improve their performance. For example, experimental results on our dataset show that with a certain integrating mode, the improvement of using DG-CNN over a baseline network structure ResNet18 is 2.17% in accuracy, 1.69% in sensitivity, 2.64% in specificity and 2.57% in AUC (Area Under Curve). To the best of our knowledge, this is the first time that the margin information is utilized to improve the performance of deep neural networks in diagnosing breast cancer in BUS images.
Document-level machine translation (MT) remains challenging due to its difficulty in efficiently using document-level global context for translation. In this paper, we propose a hierarchical model to learn the global context for document-level neural machine translation (NMT). This is done through a sentence encoder to capture intra-sentence dependencies and a document encoder to model document-level inter-sentence consistency and coherence. With this hierarchical architecture, we feedback the extracted document-level global context to each word in a top-down fashion to distinguish different translations of a word according to its specific surrounding context. Notably, we explore the effect of three popular attention functions during the information backward-distribution phase to take a deep look into the global context information distribution of our model. In addition, since large-scale in-domain document-level parallel corpora are usually unavailable, we use a two-step training strategy to take advantage of a large-scale corpus with out-of-domain parallel sentence pairs and a small-scale corpus with in-domain parallel document pairs to achieve the domain adaptability. Experimental results of our model on Chinese-English and English-German corpora significantly improve the Transformer baseline by 4.5 BLEU points on average which demonstrates the effectiveness of our proposed hierarchical model in document-level NMT.
Slot filling and intent prediction are basic tasks in capturing semantic frame of human utterances. Slots and intent have strong correlation for semantic frame parsing. For each utterance, a specific intent type is generally determined with the indication information of words having slot tags (called as slot words), and in reverse the intent type decides that words of certain categories should be used to fill as slots. However, the Intent-Slot correlation is rarely modeled explicitly in existing studies, and hence may be not fully exploited. In this paper, we model Intent-Slot correlation explicitly and propose a new framework for joint intent prediction and slot filling. Firstly, we explore the effects of slot words on intent by differentiating them from the other words, and we recognize slot words by solving a sequence labeling task with the bi-directional long short-term memory (BiLSTM) model. Then, slot recognition information is introduced into attention-based intent prediction and slot filling to improve semantic results. In addition, we integrate the Slot-Gated mechanism into slot filling to model dependency of slots on intent. Finally, we obtain slot recognition, intent prediction and slot filling by training with joint optimization. Experimental results on the benchmark Air-line Travel Information System (ATIS) and Snips datasets show that our Intent-Slot correlation model achieves state-of-the-art semantic frame performance with a lightweight structure.
DNA methylation is one important epigenetic type to play a vital role in many diseases including cancers. With the development of the high-throughput sequencing technology, there is much progress to disclose the relations of DNA methylation with diseases. However, the analyses of DNA methylation data are challenging due to the missing values caused by the limitations of current techniques. While many methods have been developed to impute the missing values, these methods are mostly based on the correlations between individual samples, and thus are limited for the abnormal samples in cancers. In this study, we present a novel transfer learning based neural network to impute missing DNA methylation data, namely the TDimpute-DNAmeth method. The method learns common relations between DNA methylation from pan-cancer samples, and then fine-tunes the learned relations over each specific cancer type for imputing the missing data. Tested on 16 cancer datasets, our method was shown to outperform other commonly-used methods. Further analyses indicated that DNA methylation is related to cancer survival and thus can be used as a biomarker of cancer prognosis.
COVID-19 is a contagious infection that has severe effects on the global economy and our daily life. Accurate diagnosis of COVID-19 is of importance for consultants, patients, and radiologists. In this study, we use the deep learning network AlexNet as the backbone, and enhance it with the following two aspects: 1) adding batch normalization to help accelerate the training, reducing the internal covariance shift; 2) replacing the fully connected layer in AlexNet with three classifiers: SNN, ELM, and RVFL. Therefore, we have three novel models from the deep COVID network (DC-Net) framework, which are named DC-Net-S, DC-Net-E, and DC-Net-R, respectively. After comparison, we find the proposed DC-Net-R achieves an average accuracy of 90.91% on a private dataset (available upon email request) comprising of 296 images while the specificity reaches 96.13%, and has the best performance among all three proposed classifiers. In addition, we show that our DC-Net-R also performs much better than other existing algorithms in the literature.
The development of IP-based Internet of Things (IoT) networks would facilitate more effective end-to-end IP network architectures, but it remains a challenge. Network routing needs to be effectively addressed in the IoT environments of scarce computational and energy resources. Accordingly, the Internet Engineering Task Force (IETF) has specified the IPv6 Routing Protocol for Low Power and Lossy Network (RPL) to provide a bespoke IPv6-based routing framework for IoT networks. However, RPL comes with no Quality of Service (QoS) support which is an essential requirement for many IoT applications. The network research community has introduced a number of research proposals enhancing RPL with different QoS solutions. This paper presents a review of these proposed solutions and aims to establish a firm understanding of recent QoS developments for RPL and possible areas for future IoT routing research. The focus is on comprehending the protocol and networking properties that can affect QoS performance in RPL networks. Consideration is also given to different objective functions developed for addressing varying QoS aspects such as throughput, delay, and packet loss. RPL is also extended in a number of QoS solutions following different approaches at the MAC, network, and application layers. However, there is still a need for further developments to address effective QoS support, particularly for dynamic RPL networks.
When querying databases containing sensitive information, the privacy of individuals stored in the database has to be guaranteed. Such guarantees are provided by differentially private mechanisms which add controlled noise to the query responses. However, most such mechanisms do not take into consideration the valid range of the query being posed. Thus, noisy responses that fall outside of this range may potentially be produced. To rectify this and therefore improve the utility of the mechanism, the commonly-used Laplace distribution can be truncated to the valid range of the query and then normalized. However, such a data-dependent operation of normalization leaks additional information about the true query response, thereby violating the differential privacy guarantee. Here, we propose a new method which preserves the differential privacy guarantee through a careful determination of an appropriate scaling parameter for the Laplace distribution. We adapt the privacy guarantee in the context of the Laplace distribution to account for data-dependent normalization factors and study this guarantee for different classes of range constraint configurations. We provide derivations of the optimal scaling parameter (i.e., the minimal value that preserves differential privacy) for each class or provide an approximation thereof. As a result of this work, one can use the Laplace distribution to answer queries in a range-adherent and differentially private manner. To demonstrate the benefits of our proposed method of normalization, we present an experimental comparison against other range-adherent mechanisms. We show that our proposed approach is able to provide improved utility over the alternative mechanisms.
With the growing requirements of web applications, web components are developed to package the implementation of commonly-used features for reuse. In some cases, the developer may want to reuse some features which cannot be customized by the component's APIs. He/she has to extract the implementation by hand. It is labor-intensive and error-prone. Considering the widely-used test cases which can be useful to specify the software features, a test-driven approach is proposed to extract the implementation of the desired features in web components. The satisfaction of the user's requirements is transformed into the passing rate of user-specified test cases. In this way, the quality of the extraction result can be evaluated automatically. Meanwhile, a record/replay-based GUI test generation method is proposed to ensure that the extraction result has the correct GUI appearance. To extract the feature implementation, a hierarchical genetic algorithm is proposed to find the code snippet that can pass all the tests and has the approximate smallest size. We compare our method with two existing feature extraction methods. The result shows that our method can extract the correct implementation with the minimum size. A human-subject study is conducted to show the effectiveness and weaknesses of our method in helping users extract the features.
Grey-box fuzzing is an effective technology to detect software vulnerabilities, such as memory corruption. Previous fuzzers in detecting memory corruption bugs either use heavy-weight analysis, or use techniques which are not customized for memory corruption detection. In this paper, we propose a novel memory bug guided fuzzer, ovAFLow. To begin with, we broaden the memory corruption targets where we frequently identify bugs. Next, ovAFLow utilizes light-weight and effective methods to build connections between the fuzzing inputs and these corruption targets. Based on the connection results, ovAFLow uses customized techniques to direct the fuzzing process closer to memory corruption. We evaluate ovAFLow against state-of-the-art fuzzers, including AFL (american fuzzy lop), AFLFast, FairFuzz, QSYM, Angora, TIFF, and TortoiseFuzz. The evaluation results show better vulnerability detection ability of ovAFLow, and the performance overhead is acceptable. Moreover, we identify 12 new memory corruption bugs and two CVEs (common vulnerability exposures) with the help of ovAFLow.
File entropy is one of the major indicators of crypto-ransomware because the encryption by ransomware increases the randomness of file contents. However, entropy-based ransomware detection has certain limitations; for example, when distinguishing ransomware-encrypted files from normal files with inherently high-level entropy, misclassification is very possible. In addition, the entropy evaluation cost for an entire file renders entropy-based detection impractical for large files. In this paper, we propose two indicators based on byte frequency for use in ransomware detection; these are termed EntropySA and DistSA, and both consider the interesting characteristics of certain file subareas termed "sample areas'' (SAs). For an encrypted file, both the sampled area and the whole file exhibit high-level randomness, but for a plain file, the sampled area embeds informative structures such as a file header and thus exhibits relatively low-level randomness even though the entire file exhibits high-level randomness. EntropySA and DistSA use "byte frequency" and a variation of byte frequency, respectively, derived from sampled areas. Both indicators cause less overhead than other entropy-based detection methods, as experimentally proven using realistic ransomware samples. To evaluate the effectiveness and feasibility of our indicators, we also employ three expensive but elaborate classification models (neural network, support vector machine and threshold-based approaches). Using these models, our experimental indicators yielded an average F1-measure of 0.994 and an average detection rate of 99.46% for file encryption attacks by realistic ransomware samples.
Oblivious polynomial evaluation (OPE) is a two-party protocol that allows a receiver, R to learn an evaluation f(α), of a sender, S's polynomial f(x), whilst keeping both α and f(x) private. This protocol has attracted a lot of attention recently, as it has wide ranging applications in the field of cryptography.
In this article we review some of these applications and, additionally, take an in-depth look at the special case of information theoretic OPE. Specifically, we provide a current and critical review of the existing information theoretic OPE protocols in the literature. We divide these protocols into two distinct cases (three-party and distributed OPE) allowing for the easy distinction and classification of future information theoretic OPE protocols. In addition to this work, we also develop several modifications and extensions to existing schemes, resulting in increased security, flexibility and efficiency. Lastly, we also identify a security flaw in a previously published OPE scheme.
In 1982, Goldwasser and Micali proposed the first probabilistic public key cryptosystem with indistinguishability under chosen plaintext attack security based on the quadratic residuosity assumption. Ciphertext expansion of Goldwasser's scheme is quite large, thereby the scheme is inefficient. A lot of schemes have been proposed to reduce the ciphertext expansion. Some schemes use the same encryption algorithm as Goldwasser's scheme with different parameters and keys, which we call them Goldwasser and Micali's type (GM-type) schemes. GM-type schemes can be divided into two categories according to different parameters and decryption algorithms. In this paper, we propose the first generalized GM-type scheme combining these two categories. All GM-type schemes are special cases of our generalized GM-type scheme. The ciphertext expansion of our scheme is smaller than that of any other GM-type schemes.
Nowadays, application migration becomes more and more attractive.
For example, it can make computation closer to data sources or make service closer to end-users,
which may significantly decrease latency in edge computing.
Yet, migrating applications among servers that are controlled by different platform owners raises security issues.
We leverage hardware-secured trusted execution environment (TEE, aka., enclave) technologies,
such as Intel SGX, AMD SEV, and ARM TrustZone, for protecting critical computations on untrusted servers.
However, these hardware TEEs propose non-uniform programming abstractions and
are based on heterogeneous architectures,
which not only forces programmers to develop secure applications targeting some specific abstraction
but also hinders the migration of protected applications.
Therefore, we propose UniTEE which gives a unified enclave programming abstraction
across the above three hardware TEEs by using a microkernel-based design
and enables the secure enclave migration by integrating heterogeneous migration techniques.
We have implemented the prototype on real machines.
The evaluation results show the migration support incurs nearly-zero runtime overhead
and the migration procedure is also efficient.
Loop and Catmull-Clark are the most famous approximation subdivision schemes, but their limit surfaces do not interpolate the vertices of the given mesh. Progressive-iterative approximation (PIA) is an efficient method for data interpolation and has a wide range of applications in many fields such as subdivision surface fitting, parametric curve and surface fitting among others. However, the convergence rate of classical PIA is slow. In this paper, we present a new and fast PIA format for constructing interpolation subdivision surface that interpolates the vertices of a mesh with arbitrary topology. The proposed method, named Conjugate-Gradient Progressive-Iterative Approximation (CG-PIA), is based on the Conjugate-Gradient Iterative algorithm and the Progressive Iterative Approximation (PIA) algorithm. The method is presented using Loop and Catmull-Clark subdivision surfaces. CG-PIA preserves the features of the classical PIA method, such as the advantages of both the local and global scheme and resemblance with the given mesh. Moreover, CG-PIA has the following features. 1) It has a faster convergence rate compared with the classical PIA and W-PIA. 2) CG-PIA avoids the selection of weights compared with W-PIA. 3) CG-PIA does not need to modify the subdivision schemes compared with other methods with fairness measure. Numerous examples for Loop and Catmull-Clark subdivision surfaces are provided in this paper to demonstrate the efficiency and effectiveness of CG-PIA.