This research paper explores the concept of edge computing and the advancements in this technology. Edge computing is a distributed computing paradigm that brings computation and data storage closer to the devices that generate and collect data in order to reduce latency and improve performance. It has been identified as a powerful tool for providing low-latency access to data and applications and is becoming increasingly popular in the IoT and 5G era. This paper focuses on the various advancements in edge computing technology, such as the emergence of edge cloud platforms and the development of edge analytics and machine learning solutions. It also discusses various applications of edge com- puting, such as autonomous driving, augmented/virtual reality, and the Internet of Things, and how they are being used to improve efficiency and reduce latency. Finally, the paper analyses the various challenges and opportunities associated with edge computing, including security, scalability and privacy concerns.

Speech COVID-19 detection systems have gained popularity as they represent an easy-to-use and low-cost solution that is well suited for at-home long-term monitoring of patients with persistent symptoms. Recently, however, the limited generalization capability of existing deep neural network based systems to unseen datasets has been raised as a serious concern, as has their limited interpretability. In this paper, we propose two innovations to help overcome these issues. First, we propose the use of a 3-dimensional modulation frequency tensor (called modulation tensorgram representation, MTR) as input to a convolutional recurrent neural network for COVID-19 detection. The representation is known to provide robustness against different environmental factors seen across datasets. Next, we propose the use of spectro-temporal saliency masking to aggregate regions of the MTR related to COVID-19, thus helping further improve the generalizability and interpretability of the model. Experiments are conducted on three public datasets and results show the proposed solution consistently outperforming two benchmark systems in within-, across-, and unseen-dataset tests. The proposed method relies on a similar number of parameters to the benchmark, thus a promising solution for at-home monitoring of COVID-19 infection.

Little is currently known about how varied source locations affect a listener’s emotional reaction to music. Here, using spectral features extracted from electrophysiology (EEG) data, we tested through machine learning whether four music source positions (front, back, left, and right) could be accurately distinguished according to the type of valence in a subject-wise manner. The findings demonstrate that distinct EEG correlates can reliably classify the four source locations and that the effect is stronger when music with a negative emotional valence is played outside of the listener’s visual field. This proof-of-concept study may pave the way for advanced spatial audio analysis approaches in music information retrieval by considering the listener’s emotional impact depending on the source direction of incidence.Final published link:

A document by Siyue Li. Click on the document to view its contents.

In this article, I/We report a novel eye-movement assessment method using a digital camera to measure eye conjugacy in healthy individuals while performing a neurological examination. This is clinically significant because this approach overcomes the limitations of complex and expensive setups (e.g., infrared cameras) that often make it impractical to scale up and translate to clinical use. Moreover, this approach removes the need for a calibration procedure which has caused prior studies to exclude participants, potentially introducing selection bias and limiting generalizability. Our study suggests that this technology could be deployed for clinical use in the clinic or pre-hospital setting, including telemedicine or emergency medical services (EMS) encounters to detect neurological injury or diseases that cause neuro-ocular deficits, like stroke.

Full leverage of the huge volume of data generated on a large number of user devices for providing intelligent services in the 6G network calls for Ubiquitous Intelligence (UI). A key to developing UI lies in the involvement of the large number of network devices, which contribute their data to collaborative Machine Learning (ML) and provide their computational resources to support the learning process. Federated Learning (FL) is a new ML method that enables data owners to collaborate in model training without exposing private data, which allows user devices to contribute their data to developing UI. Edge computing deploys cloud-like capabilities at the network edge, which enables network devices to offer their computational resources for supporting FL. Therefore, a combination of FL and edge computing may greatly facilitate the development of ubiquitous intelligence. In this article, we present a comprehensive survey of the recent developments in technologies for combining FL and edge computing with a holistic vision across the fields of FL and edge computing. We conduct our survey from both the perspective of an FL framework deployed in an edge computing environment (FL in Edge) and the perspective of an edge computing system providing a platform for supporting FL (Edge for FL). From the FL in Edge perspective, we first identify the main challenges to FL in edge computing and then survey the representative technical strategies for addressing the challenges. From the Edge for FL perspective, we first analyze the key requirements for edge computing to support FL and then review the recent advances in edge computing technologies that may be exploited to meet the requirements. Then we discuss open problems and identify some possible directions for future research on combining FL and edge computing, with the hope of arousing the research community’s interest in this emerging and exciting interdisciplinary field.

This paper studies the joint trajectory and resource allocation problem for a laser-powered unmanned aerial vehicle (UAV) assisted data aggregation framework. The considered system incorporates semi-passive nodes that establish their wireless communication links with the UAV via bistatic backscatter communications enabled by a battery-powered power beacon source. We aim to optimize the UAV trajectory while minimizing the laser energy consumption throughout the whole flight by tuning the laser power and the power beacon radiated temporal power profiles. Towards this aim, we first adopt path discretization to approximate the optimal control problem of interest into a non-linear programming one whilst accounting for the UAV dynamics constraints and available power budget restrictions. Then,  we solve the problem by successive convex approximation (SCA) over the joint set of variables and determine the problem feasibility by a data collection maximization problem. Moreover, we propose a low-complexity solution for the feasibility problem. Finally, the conducted simulations show that the proposed algorithm provides  50% collected data increase  and almost 50%  laser energy reduction  under different operation conditions such as laser station position, maximum laser output power, start and end points of the trajectory, power beacon battery capacity, and the  permissible flight length.

Small-scale ocean fronts play a significant role in absorbing the excess heat and CO2 generated by climate change, yet their dynamics are not well understood. Existing in-situ and remote sensing measurements of the ocean have inadequate spatial and temporal coverage to map small-scale ocean fronts globally. Additionally, conventional algorithms to generate ocean front maps are computationally intensive and require data with long lead times. We propose machine learning (ML) models to detect temperature and chlorophyll ocean fronts from unprocessed and radiometrically uncorrected satellite im- agery by transfer learning from existing models for edge detection. We use two separate datasets: one based on conventional approaches to ocean front detection, and a second based on human annotated ground truth1. The deep learning front detection approach significantly reduces the resources and overall lead times needed for detecting ocean fronts. The deep learning models are developed with resource-constrained edge compute platforms like CubeSats in mind, as such platforms can address the spatial and temporal coverage challenges. The highest performing models achieve accuracies of 96% and make predictions in milliseconds using unoptimized desktop CPUs and using less than 100 MB of storage; these capabilities are well- suited for CubeSat deployment.

Satellite images captured by the sensor do not contain the reflectance values returned from the earth due to absorption and scattering caused by the atmospheric elements. Atmospheric correction (AC) retrieves the actual reflectance values, called surface reflectance (SR). Existing models are based on physics-based radiative transfer codes that rely on precomputed lookup tables, require many atmospheric parameters, and involve high computation costs. A few of these parameters are difficult to calculate and must be estimated. Deep learning models can be an excellent approximator to such physics-based AC models. In the proposed paper, we design, develop and analyze a deep learning end-to-end model. It is trained with a seasonally and spatially rich dataset to perform AC of Landsat 8 satellite images without explicitly considering atmospheric parameters. Experiments on model predictions are carried out to validate the effectiveness using Landsat data and ground measurements provided by RadCalNet. The results are encouraging because the model predicts SR values with good accuracy and establishes a high correlation with the reference data.

In this paper, a computationally efficient chance-constrained rollover-free motion planning method is presented. Specifically, the method is developed to plan motions for reconfigurable vehicles with the knowledge of a 3-D terrain model that has limited accuracy. The overall motion planning problem is formulated as a nonlinear optimal control problem (NOCP) that employs a constraint in the form of a bound on the probability of rollover under terrain-induced vehicle orientation uncertainty. To increase the computational efficiency of the NOCP, a geometric interpretation of the chance constraint is derived based on the characteristics of SO(3), the 3-D rotation group. Monte Carlo simulations are provided to demonstrate the usefulness of the geometric interpretation through comparisons with other methods. Experimental data gathered from driving a mobile robot through real forests are also used to validate the proposed model. Finally, path and trajectory generation results obtained with the proposed planning method for a feller-buncher machine traversing through uncertain 3-D terrain are presented to showcase the method’s overall performance and efficiency.

Machine learning (ML) solutions are being applied in many areas of our daily lives, but they often require high-quality, balanced datasets in order to perform well. However, datasets for real-world problems are often imbalanced, requiring the use of special-purpose ML algorithms or synthetic data to address the class imbalance. Traditional techniques such as Synthetic Minority Oversampling Technique (SMOTE) and generative models such as Variational Auto Encoders (VAE) and Generative Adversarial Networks (GAN) are commonly used to generate minority class samples. Evaluating the quality of synthetic samples can be challenging, and researchers often rely on improved classifier performance as justification for their use. However, simply performing well on a test set is not sufficient to ensure the trustworthiness of a model, and further analysis of model predictions is necessary. To address this, we trained multiple classifiers on synthetic data generated by various methods and analyzed their predictions using SHAP, an explainable AI technique. Our in-depth analysis showed that these classifiers used different features for making predictions and placed different levels of importance on commonly used features. Therefore, we conclude that classification models trained on synthetic data must be carefully analyzed by human experts before being deployed in the real world.

Classification, segmentation, and the identification of the infection region in MRI images of brain tumors are labor-intensive and iterative processes. Numerous anatomical structures of the human body may be envisioned using an image processing theory. With basic imaging methods, it is challenging to see the aberrant human brain’s structure. The neurological structure of the human brain may be distinguished and made clearer using the magnetic resonance imaging technique. The MRI approach uses a number of imaging techniques to evaluate and record the human brain’s interior features. In this study, we focused on strategies for noise removal, gray-level co-occurrence matrix (GLCM) extraction of features, and segmentation of brain tumor regions based on Discrete Wavelet Transform (DWT) to minimize complexity and enhance performance. In turn, this reduces any noise that could have been left over after segmentation due to morphological filtering. Brain MRI scans were utilized to test the accuracy of the classification and the location of the tumor using probabilistic neural network classifiers. The classifier’s accuracy and position detection were tested using MRI brain imaging. The efficiency of the suggested approach is demonstrated by experimental findings, which showed that normal and diseased tissues could be distinguished from one another from brain MRI scans with about 100% accuracy.

By extending metric space to expression in hash of directed graph, spatiotemporal observation could be described in matrix quantized state. (Working paper, more contents would be added soon)

A document by Taissir Elganimi . Click on the document to view its contents.

Abstract— This work-in-progress presents a compact toolkit for reproducing the training of T5-based multiinstruments and multi-track transcription models on multiple MIR tasks. We plan to release our code in early 2023 at https://github.com/mimbres/YourMT3.

The emerging cybersecurity industry in Canada is not seen as anywhere near the scale of its American cyber-industrial complex counterpart, with limited-scale projects occurring to improve security sensor technology for institutions. However, due to international trade increasingly occurring through cyberspace, it is expected that the cyber industry will grow in Canada through both international and domestic partnerships, with Canadian critical infrastructure partners requiring greater security controls in order to ensure the confidentiality, integrity and availability of their data and operations.

Realistic Virtual Reality is supported by 360° video, which provides viewers with an immersive watching experience. However, 360° video is bulky in size, while the transmission system has limited ability to provide bandwidth. As such, intelligent adaptive delivery solutions play a crucial role in enabling users to stream high-quality 360° video. In this research, we propose a novel approach for 360- degree video streaming over HTTP/2 that can provide consumers with a good watching experience (QoE) even in varying network circumstances and head-eye movements over time. The proposed method deploys the so-called BBAG algorithm (Buffer and Bandwidth Allocation Algorithm) using Scalable Video Coding to choose appropriate tile layers to resolve the trade-off between network and user adaptivity. With the support of HTTP/2’s stream termination capability, the delivery of late tile layers is terminated to handle abrupt interruptions. Our BBAG method is proven to improve QoE from 3.56% to 93.71% and the average viewport bitrate from 1.78% to 80.49% compared to state-of-the-art methods.

In this article, the current status of a novel mi- crowave quantum radar (MQR) based on quantum illumination (QI) protocol using a Josephson traveling-wave parametric am- plifier (JTWPA) as a quantum source of broadband microwave entangled photons for detecting very low cross-radar section objects is described. Microwave entangled signal-idler pairs at 3.3 GHz and 3.45 GHz are generated, respectively, with a wide bandwidth equal to 3 GHz, a power gain equal to 15.3 dB, and a noise temperature close to the standard quantum limit (SQL) at cryogenic temperatures. A phase conjugate receiver is used as a joint detection scheme for entangled signal-idler pairs, and the expected performance of the scheme in terms of the corresponding receiver operating characteristic (ROC) curve is obtained. These results in terms of maximum dynamic bandwidth and ROC candidate our proposed scheme towards a long-range target detection MQR.