Deep-learning-based malware-detection models are threatened by adversarial attacks. This paper designs a robust and secure convolutional neural network (CNN) for malware classification. First, three CNNs with different pooling layers, including global average (GAP), max (GMP), spatial pyramid (SPP), proposed. Second, we designed an executable attack to construct changing the meaningless unimportant segments within Portable Executable (PE) header file. Finally, consolidate GMP-based CNN,...

Due to high flexibility and ease of deployment, Unmanned Aerial Vehicle (UAV)-enabled mobile edge computing (MEC) has recently emerged provide services for users meet the demands computing-intensive tasks at edge. However, MEC-server mounted UAV is inappropriate heavy-computation owing limitation energy supply hardware cost, which may make excessively CPU temperature. To tackle this issue, paper considers a UAV-and-basestation (BS) hybrid-assisted MEC network, where hover-fly-hover mode...

Although numerous state-of-the-art deep neural networks have recently been proposed for malware classification, effectively detecting on a large-scale sample set and identifying zero-day or new variants still pose significant challenges. To address this issue, hashing-based classification model is designed identification, including two parts: ResNet50-based hashing retrieval voting-based classification. Specifically, multiple models are developed by extracting the high-layer outputs (feature...

Although several cloud storage systems have been proposed, most of them can provide highly efficient point queries only because the key-value pairs storing mechanism. For these systems, satisfying complex multi-dimensional means scanning whole dataset, which is inefficient. In this paper, we propose a multidimensional index framework, based on Skip-list and Octree, refer to as Skip-Octree. Using randomized skip list makes hierarchical Octree structure easier implement in system. To support...

A typical stateful (resp. stateless) group key distribution (GKD) protocol is composed of a secret assignment algorithm, and join/leave rekeying algorithms stateless algorithm). Any design flaw in any these could lead to attacks on GKD protocols. We show how two recently-proposed protocols based asymmetric cryptographic primitives suffer from collusion due security flaws either or leave algorithms. variety single-user improvements number Shamir's Secret-Sharing Scheme (SSS) have been put...

Automated essay scoring (AES) gains momentum recently in English-based environment. However, the development of Chinese AES system is slow and fruitless. Many foreign students participate Proficiency Test (HSK) so a HSK automated (HSK AES) high demand. To develop an effective reliable system, this paper proposes three machine learning deep models that take essays as input. We apply Word2vec TF-IDF (term frequency-inverse document frequency) methods to extract important features from original...

With the vast popularity of IoT (Internet-of-Things), cloud computing and edge computing, botnet attacks are flourishing nowadays. Meanwhile, deep learning-powered models widely deployed to secure network applications. However, learning-based detection is a challenging problem due its extensive traffic volume, complex feature engineering lack benchmark dataset for evaluation. aim improving performance detection, this paper firstly designs extraction method by using effective payload from...

Centered around the model of ESDDM, a software evolution process based on behavior interface, and knowledge component classification retrieval evaluation, this paper furthered previous work by proposing reusability evaluation method fuzzy mathematics. The provides an effective way evaluating reusable ability certain during evolution.

The adversarial examples show the vulnerability of deep neural networks, which makes attacks widely concerned. However, most attack methods are based on image classification model. In this paper, we use Momentum Iterative Fast Gradient Sign Method (MI-FGSM), stabilize optimization and escape from poor local maxima, to generate Faster R-CNN object detector. We have made some improvements previous detection methods. best current method, Project Descent (PGD) detection, starts a random value,...

With the wide popularity of Internet-of-Things (IoT), machine learning-based malware detection systems are incapable being installed on IoT devices due to restricted computing power and resources. To bridging above gap, this paper proposes an integrated deep learning system for based features extracted from network packet, NetFlow samples. By referring models popular in natural language processing domain, 7 neural networks with attention mechanism designed both character-level word-level...

While deep learning models are widely adopted in malware detection, ResNet has been proved to be the most effective model many researches. However, existing models, including ResNet, failed detect packed with satisfactory accuracy. To solve this problem, a neural network framework by optimizing fragmented image and extracting key textual feature patterns is proposed for detection. Each into multiple slices points extraction two point locating algorithms, SIFT (Scale-Invariant Feature...

Backdoor attacks targeting the deep neural network are flourishing recently and more stealthy because attacked model behaviors normal on clean samples. A understanding of backdoor malware detection models is still missing. We first design a highly transferable attack method different convolutional networks (CNNs) for detection. Based computation most effective byte sub-sequence from samples, trigger patterns generated by training class activation mapping-based (CAM-DNN). The mapping applied...

Many tasks have been solved by deep neural networks in the field of natural language processing (NLP) and generation (NLG). However, some critical issues teaching Chinese as a foreign or second for students are still unsolved. The learning dynamic auxiliary words (le, zhe guo) seldom considered learning-powered applications. Taking three typical inputs, models designed evaluated our work. models, including BERT, BERT-WWM RoBERTa, optimized observing characteristics two corpora. experimental...