Search

Computing architecture continues the pendulum swing between centralized and distributed models – driven by technological innovation in CPU/GPU architecture, memory/storage, I/O, networking, performance, and emerging use cases. For a number of years, the most recent architecture has been the cloud-based centralized computing model, which is now shifting to a distributed edge computing model. New businesses, technologies, usage models, and new applications are driving this change. The rapid growth of IoT across all segments of society is driven by 5G, Edge, low-cost sensors, embedded SoC controllers, and new enterprise applications. The impact is more data being generated by these smart sensors, increased demand for storage, data analytics, and network capacity to move this data to adjacent nodes or cloud models. These and other parameters are driving the emergence of the Edge computing model – and there are many Edge types. However, in terms of optimisation and efficiency, these approaches may not be the best solution.

Trust-region (TR) type method, based on a quadratic model such as the trust-region subproblem (TRS) and $ p $-regularization subproblem ($p$RS), is arguably one of the most successful methods for unconstrained minimization. In this paper, we study a general regularized subproblem (named $ \rho $RS), which covers TRS and $p$RS as special cases. We derive a strong duality theorem for $ \rho $RS, and also its necessary and sufficient optimality condition under general assumptions on the regularization term. We then define the Rendl-Wolkowicz (RW) dual problem of $ \rho $RS, which is a maximization problem whose objective function is concave, and differentiable except possibly at two points. It is worth pointing out that our definition is based on an alternative derivation of the RW-dual problem for TRS. Then we propose an eigensolver-based algorithm for solving the RW-dual problem of $ \rho $RS. The algorithm is carried out by finding the smallest eigenvalue and its unit eigenvector of a certain matrix in each iteration. Finally, we present numerical results on randomly generated $p$RS's, and on a new class of regularized problem that combines TRS and $p$RS, to illustrate our algorithm.

Although SISR (Single Image Super Resolution) problem can be effectively solved by deep learning based methods, the training phase often considers single degradation type such as bicubic interpolation or Gaussian blur with fixed variance. These priori hypotheses often fail and lead to reconstruction error in real scenario. In this paper, we propose an end-to-end CNN model RPSRMD to handle SR problem in multiple Gaussian degradations by extracting and using as side information a shared image prior that is consistent in different Gaussian degradations. The shared image prior is generated by an AED network RPGen with a rationally designed loss function that contains two parts: consistency loss and validity loss. These losses supervise the training of AED to guarantee that the image priors of one image with different Gaussian blurs to be very similar. Afterwards we carefully designed a SR network, which is termed as PResNet (Prior based Residual Network) in this paper, to efficiently use the image priors and generate high quality and robust SR images when unknown Gaussian blur is presented. When we applied variant Gaussian blurs to the low resolution images, the experiments prove that our proposed RPSRMD, which includes RPGen and PResNet as two core components, is superior to many state-of-the-art SR methods that were designed and trained to handle multi-degradation.

Air Quality Index is the measure of pollutants in the atmosphere based on readings from a number of sensors. This paper uses a deep neural algorithm to predict AQI resulting in a reduced number of sensors. The method employed enables specific sensor data to be substituted or replaced, while still obtaining robust results.

peer-reviewed The potential of the second wave of Artificial Intelligence (AI) to change our lives beyond recognition is both exciting and challenging. AI has been around for over three decades, and this new approach of artificial intelligence, due to enhancements in technology, both software, and hardware, has resulted in the fact that human decision-making is considered inferior and erratic in many fields: none more so than medicine. Machine learning algorithms with access to large data sets can be trained to outperform clinicians in many respects. AI’s effectiveness in accurate diagnosis of various medical conditions and medical image interpretation is well documented. Modern AI technology has the potential to transform medicine to a level never seen before in terms of efficiency and accuracy; but is also potentially highly disruptive, creating insecurity and allowing the transfer of expert domain knowledge to machines. Anesthetics is a complex medical discipline and assuming AI can easily replace experienced and knowledgeable medical practitioners is a very unrealistic expectation. AI can be used in anesthetics to develop, in some respects, more advanced clinical decision support tools based on machine learning. This paper focuses on the complexity of both AI developments, deep learning, neural networks, etc. and opportunities of AI in anesthetics for the future. It will review current advances in AI tools and hardware technologies as well as outlining how these can be used in the field of anesthetics.

Localization and mapping in a dynamic scene is a crucial problem for the indoor visual simultaneous localization and mapping (SLAM) system. Most existed visual odometry (VO) or SLAM systems are based on the assumption that the environment is static. The performance of a SLAM system may degenerate when it is operated in a severely dynamic environment. The assumption limits the applications of RGB-D SLAM in the dynamic environment. In this paper, we propose a workflow to segment the objects accurately, which will be marked as the potentially dynamic-object area based on the semantic information. A novel approach for motion detection and removal from the moving camera is introduced. We integrate the semantics-based motion detection and the segmentation approach with an RGB-D SLAM system. To evaluate the effectiveness of the proposed approach, we conduct the experiments on the challenging dynamic sequences of TUM-RGBD datasets. The experimental results suggest that our approach improves the accuracy of localization and outperforms the state-of-the-art dynamic-removal-based SLAM system in both severely dynamic and slightly dynamic scenes.

In an era of smart technology where the human face is used for authentication because its fast and convenient but not every technology introduced is 100% credible. Face spoofing has rendered this technology vulnerable because of the damage it can cause when its security is breached. 3D face spoof has emerged and complicated facial recognition systems. Although, sophisticated improvement over the years have been made to reduce the impact of 3D face spoofing, a robust solution that can change the facial recognition sensors from being fooled are still studied and been researched on. In this paper, we proposed a (Convolutional Neural Network, CNN) architecture focused on Fusion-based approach, Depth and Patch-based (Convolutional Neural Network, CNN) by extracting from the human facial images, the facial features and complete detail hints. Our tests were carried out on CASIA-SURF dataset consisting of 3 modalities: Color, Depth and (Infrared, IR).

In this paper, a novel wideband band-pass center frequency tunable filter is proposed based on liquid crystals for applications at millimeter-waves, which uses a micro-strip periodical structure with loading square spiral resonators. The tunable range of the proposed filter is about 1.59 GHz while the external bias voltage of the liquid crystal turns from 0V to 30V. The filter center frequency ranges from 39.08 GHz to 40.67 GHz. And the insertion loss of filter is about 3.9dB. The proposed filter has many advantages, such as continuously electrically tunability, miniaturization, low tuning voltage, etc. Therefore, a great potential of such a filter has been shown for use in the next communication system.

With more attention being paid to environmental issues in recent years, air quality monitoring is becoming more important. It has been possible to monitor air quality successfully for many years, but monitoring has traditionally been both expensive and immobile, thus restricted in application. To improve urban environments, air quality monitoring has to be widespread, ubiquitous, cheap, and rapidly responsive. Good, timely data is the key to first identifying, then tackling air pollution issues. This paper develops an alternative, cheap, IoT-based air quality monitor, which can track air pollution in real time, and transmit the relevant data rapidly through a low power wide area network. A large network of such monitors can generate a vast amount of data, which may then be processed and analyzed in the cloud in real time, and correlated with time of day, month, year or weather and other factors.

Existing SISR (single image super-resolution) methods mostly assume that a low-resolution (LR) image is bicubicly down-sampled from its high-resolution (HR) counterpart, which inevitably give rise to poor performance when the degradation is out of assumption. To address this issue, we propose a framework PRED (parallel residual and encoder-decoder network) with an innovative training strategy to enhance the robustness to multiple degradations. Consequently, the network can handle spatially variant degradations, which significantly improves the practicability of the proposed method. Extensive experimental results on real LR images show that the proposed method can not only produce favorable results on multiple degradations, but also reconstruct visually plausible HR images.

Screen content videos (SCVs) are becoming popular in many applications. Compared with natural content videos (NCVs), the SCVs have different characteristics. Therefore, the screen content coding (SCC) based on HEVC adopts some new coding tools (intra block copy and palette mode etc.) to improve coding efficiency, but these tools increase the computational complexity as well. In this paper, we propose to predict the CU partition of the SCVs by a convolutional neural network (CNN) which is trained by the large-scale database that we firstly established for screen content coding. The proposed approach is implemented in SCC reference software SCM-6.1. Experimental results show that our proposed approach can save 53.2% encoding time with 2.67% BD-rate increase on average in All Intra (AI) configurations.

As an important way of human communication, facial expression not only reflects our mental activities but also provides useful information for human behavior research. Recently, 3D technology becomes promising method to achieve robust facial expression analysis. 3D face scans are robust to lighting and pose variations. In this paper, a novel deep feature fusion convolution neural network (CNN) is designed for 3D facial expression recognition (FER). Each 3D face scan is firstly represented as 2D facial attribute maps (including depth, normal, and shape index values). Then, we combine different of facial attribute maps to learn facial representations by fine-tuning a pre-trained deep feature fusion CNN subnet trained from a large-scale image dataset for universal visual tasks. Moreover, Global Average Pooling is utilized to reduce the number of parameters to decrease the effect of overfitting. The experiments are conducted on the Bosphorus database and the results demonstrate the effectiveness of the proposed method.

A microwave frequency continuously electrically tunable split ring resonator (SRR) band-pass filter using nematic liquid crystal (LC) is proposed in this paper. The nematic liquid crystal is used to obtain the frequency shift of microwave band by changing its permittivity at room temperature (20°C-2S∘C). The permittivity is controlled by the bias voltage. The simulation results show that 890 MHz offset is achieved. The proposed filter has many advantages, such as continuously electrically tunability, miniaturization, low tuning voltage, etc. Therefore, a great potential of such a filter has been shown for use in modern microwave communication applications.

In this paper, we propose a novel method for race-walking gait analysis based on RGB images. Our method provides an efficient way to quantitatively evaluate an athlete's racewalking techniques, which enables them to improve through more precise training programs accordingly. Our method includes angle prediction and pace measurement, which in combination can indicates the impact of a racewalker's unique racewalking gait on his/her competition performance in terms of time. 1) In angle prediction, the calculation about the angle of limbs is based on the coordinates of the human skeleton points. 2) In pace measurement, we derive the racewalker's pace by detecting the distance covered by the racewalker in RGB images during a period of time. Finally, the results of a practical experiment on professional racewalking athletes show that our method can efficiently quantify and analyze the racewalking gait with high precision.