Subspace clustering methods based on data self-expression have become very popular for learning from that lie in a union of low-dimensional linear subspaces. However, the applicability subspace has been limited because practical visual raw form do not necessarily such On other hand, while Convolutional Neural Network (ConvNet) demonstrated to be powerful tool extracting discriminative features data, training ConvNet usually requires large amount labeled which are unavailable applications. To...

Speaker recognition is a task that identifies the speaker from multiple audios. Recently, advances in deep learning have considerably boosted development of speech signal processing techniques. or has been widely adopted such applications as smart locks, vehicle-mounted systems, and financial services. However, neural network-based systems (SRSs) are susceptible to adversarial attacks, which fool system make wrong decisions by small perturbations, this drawn attention researchers security...

Deep neural networks (DNNs) can improve the image analysis and interpretation of remote sensing technology by extracting valuable information from images, has extensive applications such as military affairs, agriculture, environment, transportation, urban division. The DNNs for object detection identify analyze objects in images through fruitful features which improves efficiency processing enables recognition large-scale images. However, many studies have shown that deep are vulnerable to...

Based on the Fisher–Yatess scrambling and DNA coding technology, a chaotical image encryption method is proposed. First, SHA-3 algorithm used to calculate hash value of initial password, which as chaotic system. Second, sequence are scramble plaintext, sorting for secondary scrambling. Then, rules change plaintext pixel values, makes ciphertext more random resistant attacks, thus ensures that encrypted secure. Finally, we add statistics pixel-level diffusion ensure sensitivity. The...

High dimensional data often lie approximately in low subspaces corresponding to multiple classes or categories. Segmenting the high into their is referred as subspace clustering. State of art methods solve this problem two steps. First, an affinity matrix built from based on self-expressiveness model, which each point expressed a linear combination other points. Second, segmentation obtained by spectral However, solving dependent steps separately still suboptimal. In paper, we propose joint...

AbstractAbstract<title/> Background and aims: Stroke is the number one cause of death in China. Although effective management has reduced mortality lengthened survival, little attention been paid to nutritional issues patients with stroke This study aimed assess premorbid nutrition status dying acute cerebral infarction. Methods: In this study, a total 185 ischemic within 30 days were recruited from medical records. Characteristics assessed on admission, serum biochemical parameters...

Most existing deep neural networks (DNNs) are inexplicable and fragile, which can be easily deceived by carefully designed adversarial example with tiny undetectable noise. This allows attackers to cause serious consequences in many DNN-assisted scenarios without human perception. In the field of speaker recognition, attack for recognition system has been relatively mature. works focus on white-box attacks that assume information DNN is obtainable, only a few study gray-box attacks. this...

Subspace clustering refers to the problem of segmenting high dimensional data drawn from a union subspaces into respective subspaces. In some applications, partial side-information indicate "must-link" or "cannot-link" in is available. This leads task subspace with side-information. However, prior work supervision value for has not been fully exploited. To this end, paper, we present an enhanced approach constrained side-information, termed Constrained Sparse Clustering plus (CSSC+), which...

Standard methods of subspace clustering are based on self-expressiveness in the original data space, which states that a point can be expressed as linear combination other points. However, real raw form usually not well aligned with model. Therefore, it is crucial to obtain proper feature space for performing high quality clustering. Inspired by success Convolutional Neural Networks (CNN) extraction powerful features from visual and block diagonal prior learning good affinity matrix...

Live video streaming is becoming increasingly popular as a form of interaction in social applications. One its main advantages an ability to immediately create and connect community remote users on the spot. In this paper we discuss how feature can be used for crowdsourced completion simple visual search tasks (such finding specific objects libraries stores, or navigating around live events) interactions through mobile mixed reality telepresence interfaces. We present prototype application...

Adversarial examples pose a security threat to Autopilot's speech command recognition module, which attracted widespread attention from researchers. Previous works purify the malicious adversarial perturbations through pre-processing data time and frequency domain information. However, these methods either have weak purification capacity or require significant cost. To tackle this problem, we propose real-time efficient purification-based defense method DualPure, combines two aspects in for...

This paper presents EXController, a new controller-mounted finger posture recognition device specially designed for VR handheld controllers. We seek to provide additional input through real-time vision sensing by attaching near infrared (NIR) camera onto the controller. and implemented an exploratory prototype with HTC Vive The NIR is modified from traditional webcam applied data-driven Convolutional Neural Network (CNN) classifier. 12 different gestures trained CNN classifier dataset 20...

Based on the teaching reform model of University Chicago School Medicine,Wuhan Medicine firstly integrates Neuroscience course for undergraduated medical students,which combining basic theory courses nervous system with neurological clinical cases,the aim which is to reduce duplication.In order deepen students understand and master neuroscience course,the case-based method were introduced.After two years experiences,teachers all reflect good.This article introduces this used. Key words:...

Deep neural networks (DNNs) have greatly improved speaker recognition performance. However, DNNs are vulnerable to imperceptible adversarial example, which has attracted the attention in both academic and industrial areas recent years. Although many elegant attack methods been proposed, few of them studied Universal Adversarial Perturbation (UAP) against recognition. In this paper, we propose a novel algorithm from perspective dominant features generate UAP mislead DNNs. Our method uses...