Deep learning’s rapid gains in automation are making it more popular in a variety of complex jobs. The self-driving object is an emerging technology that has the potential to transform the entire planet. The steering control of an automated item is critical to ensuring a safe and secure voyage. Consequently, in this study, we developed a methodology for predicting the steering angle only by looking at the front images of a vehicle. In addition, we used an Internet of Things-based system for collecting front images and steering angles. A Raspberry Pi (RP) camera is used in conjunction with a Raspberry Pi (RP) processing unit to capture images from vehicles, and the RP processing unit is used to collect the angles associated with each image. Apart from that, we’ve made use of deep learning-based algorithms such as VGG16, ResNet-152, DenseNet-201, and Nvidia’s models, all of which were trained using labeled training data. Our models are End-to-End CNN models, which do not require extracting elements from data such as roads, lanes, or other objects before predicting steering angle. As a result of our comparative investigation, we can conclude that the Nvidia model’s performance was satisfactory, with a Mean Squared Error (MSE) value of 0. But the Nvidia model outperforms the other pre-trained models, even though other models work well.

This document pretends to provide an overview about the lights and shadows on the latest trends in this specific area. Unlike previously released literature reviews, that are providing a wide overview about any type of AI techniques applied to overall aspects of the pandemics, this document will focus specifically on the use of DL techniques applied to COVID-19 time series forecasting. The production in this field within the last months has become quite large. After setting a group of quality criteria, related to problem definition, dataset manipulation, model identification and evaluation, 96 papers has been screened. Most of the analysed papers did not meet the common quality standards of scientific work: none of them positively scored in all of the criteria, while only about one third scored positively in at least half of the defined criteria. The emergency character of this scientific production led to getting away from some of the basic requirements for quality scientific work.

Over the past few years, the use of wireless sensor networks in a range of Internet of Things (IoT) scenarios has grown in popularity. Since IoT sensor devices have restricted battery power, a proper IoT data aggregation approach is crucial to prolong the network lifetime. To this end, current approaches typically form a virtual aggregation backbone based on a connected dominating set or maximal independent set to utilize independent transmissions of dominators. However, they usually have a fairly long aggregation delay because the dominators become bottlenecks for receiving data from all dominatees. The problem of time-efficient data aggregation in multichannel duty-cycled IoT sensor networks is analyzed in this paper. We propose a novel aggregation approach, named LInk-delay-aware REinforcement (LIRE), leveraging active slots of sensors to explore a routing structure with pipeline links, then scheduling all transmissions in a bottom-up manner. The reinforcement schedule accelerates the aggregation by exploiting unused channels and time slots left off at every scheduling round. LIRE is evaluated in a variety of simulation scenarios through theoretical analysis and performance comparisons with a state-of-the-art scheme. The simulation results show that LIRE reduces more than 80% aggregation delay compared to the existing scheme.

Neural networks with partially random weights are currently not really an independent field of research. However, the first works on random neural networks date back to the 1990s and in the last three decades there have been important new works in which random weights have been used and which are promising in that they give surprisingly good results when compared to approaches in which all weights are trained. These works, however, come from very different subareas of neural networks: Random Feedforward Neural Networks, Random Recurrent Neural Networks and Random ConvNets. In this paper, I analyze the most important works from these three areas and thereby follow a chronological order. I also work out the core result of each work. As a result, the reader can get a quick overview of this field of research.

Internet of Things (IoT) with edge computation enhances efficiency, safety, and availability of an industrial automation system. However, there is a continued effort to increase the reliability of the system with minimal downtime. This can be achieved through a modular, re-configurable, and integrable system design approach. In this paper, we propose Pro-Edge, a programmable edge network to reconfigure different services associated with industrial applications and networks. Pro-Edge employs programmable layers at the edge for re-configuring the sensor/actuator network and applications. The lowermost layer allows to reconfigure the communication related parameters and the middle layer consists of a Software-Defined Network (SDN) controller that can dynamically program different modules, handling actuation decisions from the edge. An interfacing protocol between the layers is proposed to provide reliability by considering the optimal configuration parameters among the layers. As a top-layer, a priority forwarding mechanism is designed for SDN core communication in case the sensor and actuator are in different edges. Pro-Edge significantly improves the actuation-latency and is highly energy efficient compared to the existing state-of-the-art.

Diabetes is a disease that affects how your body processes blood sugar and is often referred to as diabetes mellitus. Insulin insufficiency and ineffective insulin use coincide when the pancreas cannot produce enough insulin or the human body cannot use the insulin that is produced. Insulin is a hormone produced by the pancreas that helps in the transport of glucose from food into cells for use as energy. The common effect of uncontrolled diabetes is hyper-glycemia, or high blood sugar, which plus other health concerns, raises serious health issues, majorly towards the nerves and blood vessels. According to 2014 statistics, people aged 18 or older had diabetes and, according to 2019 statistics, diabetes alone caused 1.5 million deaths. However, because of the rapid growth of machine learning(ML) and deep learning (DL) classification algorithms. indifferent sectors, like health science, it is now remarkably easy to detect diabetes in its early stages. In this experiment, we have conducted a comparative analysis of several ML and DL techniques for early diabetes disease prediction. Additionally, we used a diabetes dataset from the UCI repository that has 17 attributes, including class, and evaluated the performance of all proposed machine learning and deep learning classification algorithms using a variety of performance metrics. According to our experiments, the XGBoost classifier outperformed the rest of the algorithms by approximately 100.0%, while the rest of the algorithms were over 90.0% accurate.

Laplacian kernels, which are widely used as sharpening filters in image processing, are isotropic and tend to over-highlight fine details with a sharp discontinuity in images. To address this issue, this paper introduces a method that integrates anisotropic averaging with the Laplacian kernels. The proposed method can also be useful as a new type of image convolution for designing convolutional neural networks.

Due to demand for information ubiquity and large-scale automation, proliferating Internet-connected heterogeneous devices exhibit significant variations in data processing capacities, purposes, operating principles, underlying protocols, and dynamic contexts. As a result, adversarial entities exploit the increasing heterogeneous network (HetIoT) vulnerabilities, leading to frequent high-impact attacks due to anomalous device interactions and scarce knowledgebase. This paper presents a two-fold solution to the problem through a network intrusion detection and prevention framework for HetIoT, called \textit{HetIoT-NIDPS}. Firstly, we assign fault scores to the Expert-curated Knowledgebase (EK) framework, correlating with low-level alerts to assess threat severity and achieve context-awareness. Secondly, the proposed Beta distribution-based HetIoT traffic behavior approximation facilitates class imbalance invariance and improves classifier performance. Additionally, the HetIoT-NIDPS can detect zero-day attacks by identifying known attack variations upon encountering unseen traffic instances. Furthermore, the dynamic HetIoT contexts necessitate real-time threat assessment through online training—performed by analyzing small batches of network traffic samples. We propound the \textit{CorrELM} classifier based on the extreme learning machine algorithm and test the hypotheses on the Bot-IoT dataset. Finally, we prioritize the correlated alerts based on their severity, determined from root cause analysis and threat severity assessment tables. The results obtained prove that the proposed HetIoT-NIDPS framework is context-aware—producing reduced false alerts, class imbalance invariant—facilitating near real-time threat assessment with unbiased classifier performance, and generalizable—applicable to many NID datasets, which the existing techniques lack when combined.

“This work has been submitted to the IEEE Transactions on Affective Computing for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.” Physiological sensing has long been an indispensable fixture for virtual reality (VR) gaming studies. Moreover, VR induced stressors are increasingly being used to assess the impact of stress on an individual’s health and well-being. This study discusses the results of experimental research comprising multimodal physiological signal acquisition from 31 participants during a Go/No-Go VR-based shooting exercise where participants had to shoot the enemy and spare the friendly targets. The study encompasses multiple sessions, including orientation, thresholding, and shooting. The shooting sessions consist of tasks under low & high difficulty induced stress conditions with in-between baseline segments. Machine learning (ML) performance with heart rate variability (HRV) from electrocardiogram (ECG) and electroencephalogram (EEG) features outperform the prevalent methods for four different VR gaming difficulty-induced stress (GDIS) classification problems (CPs). Further, the significance of the HRV predictors and different brain region activations from EEG is deciphered using statistical hypothesis testing (SHT). The ablation study shows the efficacy of multimodal physiological sensing for different gaming difficulty-induced stress classification problems (GDISCPs) in a VR shooting task.

Unmanned aerial vehicles (UAVs) can be integrated into wireless sensor networks (WSNs) for smart city applications in several ways. Among them, a UAV can be employed as a relay in a “store-carry and forward” fashion by uploading data from ground sensors and metering devices and, then, downloading it to a central unit. However, both the uploading and downloading phases can be prone to potential threats and attacks. As a legacy from traditional wireless networks, the jamming attack is still one of the major and serious threats to UAV-aided communications, especially when also the jammer is mobile, e.g., it is mounted on a UAV or inside a terrestrial vehicle. In this paper, we investigate anti-jamming communications for UAV-aided WSNs operating over doubly-selective channels in the downloading phase. In such a scenario, the signals transmitted by the UAV and the malicious mobile jammer undergo both time dispersion due to multipath propagation effects and frequency dispersion caused by their mobility. To suppress high-power jamming signals, we propose a blind physical-layer technique that jointly detects the UAV and jammer symbols through serial disturbance cancellation based on symbol-level post-sorting of the detector output. Amplitudes, phases, time delays, and Doppler shifts – required to implement the proposed detection strategy – are blindly estimated from data through the use of algorithms that exploit the almost-cyclostationarity properties of the received signal and the detailed structure of multicarrier modulation format. Simulation results corroborate the anti-jamming capabilities of the proposed method, for different mobility scenarios of the jammer.

The radar sounder data (radargrams) are used in this research work for subsurface target characterizations. We develop a hybrid Transformer-based Deep Learning framework in the domain of semantic segmentation of radar sounder data.

Zero common mode voltage (ZCMV) space vector modulation (SVM) strategy applied to a three-phase multilevel inverter (MLI) eliminates the common mode voltage (CMV). However, the usage of ZCMV-SVM strategy reduces the number of levels in the output voltage and requires higher magnitude dc voltage source due to the reduced modulation depth of the employed PWM scheme. Moreover, the usage of a single dc-source in such systems may have issues with respect to capacitor voltage balancing. Taking into account all the above issues, a 3-level inverter solution is proposed in this manuscript. The complete details of the method used for developing the proposed solution using the ZCMV space vectors is also included in this paper. The proposed topology utilizes a unique combination of a T-type 3-level inverter and a switched-capacitor (SC) circuit to achieve ZCMV performance with a single dc-source of low magnitude at a reduced component count. Analysis of the CMV and terminal voltages along with the design of the SCs are presented in this paper. The proposed topology is compared with the existing topologies to prove its unique merits over the other 3LI solutions with ZCMV capability. All the claims are validated using simulation and experimental results.

Information-preservation is recognized as the only principle for probability-possibility transformation in this work and the normalized transformation is the right method. This is based on the viewpoint that the reason we can transfer probability and possibility is that we believe the uncertainty being handled can be information-equivalently described by both probability and possibility. That viewpoint is endorsed by the random-fuzzy dual interpretation of unknown uncertainty, which says that unknown uncertainty being estimated could be interpreted as either randomness or fuzziness, depending on the available prior information and the perspective of cognition and modeling. Information of uncertain variable is defined in this work as its distribution. The suggested information-preservation principle is different from Klir's principle, which is in fact an entropy-preservation principle. Then we investigated the problem of information preservation and propagation in parallel probability-possibility systems. By parallel, we mean the two uncertainty systems have the same priori information. After uncertainty propagation the two parallel systems will generally bifurcate, which means information preservation only holds locally between the two parallel systems. This observation accords with our intuition since probability and possibility use different normalizations as well as different disjunctive operators, which makes them two different uncertainty systems appropriate for randomness and fuzziness, respectively.

The paper describes ML techniques that can be leveraged to assist signal detection in ambient backscatter communication networks.

we characterized accumulated body devices in the 100 K to 400 K range, observed the performance characteristics, and explored the ability to dynamically tune Vt as the temperature is changed and the temperature range these devices can reliably operate at.

This paper proposes an integrated magnetics, insulation, and cooling architecture to improve the thermal performance of a high frequency permanent magnet (PM) motor. The proposed architecture can be used for any motor topology to improve its thermal and insulation performance. The proposed stator yoke design interleaves copper sheets between yoke core lamination to achieve better thermal conduction from winding to heat sink. A ceramic winding holder is integrated into the armature to introduce a parallel thermal conduction path from windings to the iron yoke and to provide additional insulation. The architecture is applied to a 300 kW slotless PM synchronous motor consisting of an outer rotor Halbach PM array, slotless stator, and heatsink. 3D electromagnetic finite element methods (FEM), 2D heat transfer FEM, and an analytical thermal circuit are used to analyze the architectures impact on torque production, eddy currents, and thermal performance when compared to the baseline motor. Finally, a pole-pair prototype was built as a proof-of-concept and to verify the performance benefits of the proposed architecture.

A document by Krzysztof Michalski . Click on the document to view its contents.

Using data augmentation techniques, unsupervised representation learning methods extract features from data by training artificial neural networks to recognize that different views of an object are just different instances of the same object. We extend current unsupervised representation learning methods to networks that can self-organize data representations into two-dimensional (2D) maps. The proposed method combines ideas from Kohonen’s original self-organizing maps (SOM) and recent development in unsupervised representation learning. A ResNet backbone with an added 2D Softmax output layer is used to organize the data representations. A new loss function with linear complexity is proposed to enforce SOM requirements of winner-take-all (WTA) and competition between neurons while explicitly avoiding collapse into trivial solutions. We show that enforcing SOM topological neighborhood requirement can be achieved by a fixed radial convolution at the 2D output layer without having to resort to actual radial activation functions which prevented the original SOM algorithm from being extended to nowadays neural network architectures. We demonstrate that when combined with data augmentation techniques, self-organization is a simple emergent property of the 2D output layer because of neighborhood recruitment combined with WTA competition between neurons. The proposed methodology is demonstrated on SVHN and CIFAR10 data sets. The proposed algorithm is the first end-to-end unsupervised learning method that combines data self-organization and visualization as integral parts of unsupervised representation learning.