We present a novel adaptive mesh refinement algorithm for efficiently solving a continuous Galerkin formulation of the 2D Maxwell Eigenvalue Problem over quadrilateral discretizations. The algorithm harnesses a Refinement-by-Superposition framework with support for anisotropic hp-adaptivity. Following a brief summary of the underlying methodology, we develop a novel approach to directing and deploying intelligent fully anisotropic hp-refinements. Numerical examples targeting the Maxwell Eigenvalue problem verify the ability to achieve exponential rates of convergence on a challenging benchmark and demonstrate that including anisotropic refinement yields a significant enhancement of computational efficiency. The success of this algorithm, along with the simplicity of the underlying Refinement-by-Superposition approach, creates a promising path for implementing highly efficient yet lightweight finite element method (FEM) codes for a variety of applications in computational electromagnetics (CEM) and elsewhere.

Active second-harmonic (2ω) filters have become an integral technology for single-phase power converters in high power density designs. The series-stacked buffer (SSB) has emerged as an attractive topology among the existing solutions due to its low power, high efficiency, and compact design. However, one of the challenges in adopting SSB lies in its control, where conventionally hysteresis current control has been adopted. This results in a wide variation in the switching frequency, making the digital control implementation and filter design complex. On the other hand, fixed-frequency pulse-width modulation (PWM) control necessitates developing a model for the systematic controller design to ensure stability and desired filtering performance. The assumption of SSB modelled as a second-harmonic current source, independent of the dc bus circuit components, fails to capture the dc bus loading effect on the SSB. In this work, a dynamic model of SSB is eveloped where the main dc bus and its passive elements are taken into account. The derived model gives a systematic procedure for the controller design while ensuring the desired 2ω filtering. The stability limits and parameter sensitivities of the model are studied analytically and in simulations. The analytical model is validated, and the proposed controller performance and stability limit are verified experimentally on a hardware prototype.

Thermal management is the current bottleneck in advancement of high-power integrated circuits (ICs), and phase change heat sinks are a promising solution. With a unique structural configuration consisting of a membrane positioned above the heater surface, membrane-based heat sinks (MHS) have thus far attained heat fluxes of up to 2 kW/cm2 and HTC of up to 1.8 MW/m2K using water as the working fluid. This work reports the latest progress and performance evaluation of MHS for high flux thermal management. MHS is implemented in conjunction with a low surface tension liquid to rapidly expel bubbles from the heated surface and reach a CHF of 340 W/cm2 and a HTC of 120 kW/m2K. A parametric comparison shows that thermal efficiency, defined as the ratio of cooling capacity and pumping power consumption, of the prototypical devices exceed values reported hitherto in literature by more than two orders of magnitude. Our results indicate that coupled with surfaces of higher thermal conductivity and membranes of higher permeability, MHS devices could be a promising solution to thermal management needs of high-power electronics and lasers.

There has been a lot of hype about machine learning. The advent of big data, cloud computing, and the Internet of Things, together with powerful ML tools, have unleashed a wave of excitement around the benefits of learning. Unfortunately, the results have not always been that impressive. We need better algorithms, better data, more effective models, more powerful computing, and faster and smarter data access to get the most from these techniques. Here we present five challenges to the future of ML. As the AI and big data industries grow, we are witnessing an increasing demand for ML engineers. Machine Learning is fast becoming a core skill for every digital leader and will remain so for the foreseeable future. ML can transform data into knowledge. This is something organizations need to create insights from data. In this paper, we discuss the potential of using machine learning and big data to predict the future of intelligence. We take inspiration from the many projects in the field that are already making excellent progress and explore the possibility of creating a large repository of knowledge which could be used to build an AI machine capable of self-learning and thus be forever evolving. We explore the question of what the future might hold for artificial intelligence and machine learning, and whether it could one day replace us.

With the surge of deep learning framework, various studies have attempted to address the challenges of sentiment analysis of tweets (data sparsity, under-specificity, noise, and multilingual content) through text and network-based representation learning approaches. However, limited studies on combining the benefits of textual and structural (graph) representations for sentiment analysis of tweets have been carried out. This study proposes a multi-view learning framework (end-to-end and ensemble-based) that leverages both text-based and graph-based representation learning approaches to enrich the tweet representation for sentiment classification. The efficacy of the proposed framework is evaluated over three datasets using suitable baseline counterparts. From various experimental studies, it is observed that combining both textual and structural views can achieve better performance of sentiment classification tasks than its counterparts.

In traditional machine learning environments, the use of non-parametric error estimation to set the discriminative threshold of a classifier to achieve the best accuracy is very effective. This method is not effective in a transfer learning environment because it is only reliable when both the training and testing data have similar distributions which is not the case in a transfer learning setting. Although the use of control variate techniques has been proposed to exploit the information about the error in the training sample to reduce the error in the test sample, this method yields a finite variance and the model uncertainty is not distributed among the variance. In this paper, we proposed and test a new transfer learning validation method called control linear minimum mean-squared error (CLMMSE) for source model selection under homogeneous transfer learning settings with the absence of an adequate pre-trained source model. Our approach adopts the Bayesian linear minimum mean-squared error (LMMSE) and integrates the idea of importance sampling into a control variate approach to provide an accurate estimate for the LMMSE that is then used to select the optimal source model. By combining importance sampling with the control variate technique to reduce further the variance, we can achieve a much tighter bound with the LMMSE. This approach reduces the risk in the target domain under data shift. Experimental results on synthetic data under two data shift settings demonstrate the efficacy of our approach. A further experiment on two real-world datasets shows that we were able to improve the accuracy of two state-of-the-art models tested; Bert (0.94\% to 65\%) and CodeBERT (1.82\% to 18.2\%) when compared to using previous selection methods.

A medida que avanza la edad, pueden aparecer algunas enfermedades que poco a poco impactan en la salud de las personas. Entre ellas, algunas enfermedades crónicas y neurodegenerativas que impiden mantener la independencia en las actividades básicas de la vida diaria, aumentando los riesgos en el hogar, como las caídas que, además de impactar en la autonomía, provocan una disminución del bienestar y la calidad del estilo de vida. El actual avance de las tecnologías de la información y el abaratamiento del hardware permiten afrontar este problema desde una perspectiva preventiva más que reactiva, es decir antes de que las consecuencias para la salud se vuelvan irremediables. Las viviendas inteligentes surgen desde esta perspectiva, definidas como ambientes confortables que cuentan con inteligencia ambiental y control automático, que permiten responder al comportamiento de los residentes de manera no intrusiva ni invasiva. Este trabajo muestra parte de los resultados de una intervención de prueba clínica que duró 10 meses y donde se implementó un sistema de monitoreo denominado Senior Guardian. Este cuenta con alertas de emergencia, botón de pánico y sensores de actividad (nicturia, caídas y actividades diarias). Entre los resultados se encontró que, con la prueba de concepto, se logró obtener un sistema que permite la asistencia remota como modelo alternativo de atención a la salud de los adultos mayores. Esto demostró que brindó seguridad y protección en su entorno doméstico cotidiano, donde el 81% de los participantes mantuvo y, en su mayoría, mejoró su percepción de calidad de estilo de vida. Entre los resultados se encontró que, con la prueba de concepto, se logró obtener un sistema que permite la asistencia remota como modelo alternativo de atención a la salud de los adultos mayores. Esto demostró que brindó seguridad y protección en su entorno doméstico cotidiano, donde el 81% de los participantes mantuvo y, en su mayoría, mejoró su percepción de calidad de estilo de vida. Entre los resultados se encontró que, con la prueba de concepto, se logró obtener un sistema que permite la asistencia remota como modelo alternativo de atención a la salud de los adultos mayores. Esto demostró que brindó seguridad y protección en su entorno doméstico cotidiano, donde el 81% de los participantes mantuvo y, en su mayoría, mejoró su percepción de calidad de estilo de vida.

It is a review article on the topic of energy harvesting.

In recent years, the wearable motion capture technology has been developed rapidly in various applications. However, conventional methods usually emphasize capturing the whole body skeleton or limb movements, without considering the personalized human body data and fine-grained deformation information. Thus, it is important to develop a proper wearable motion and deformation capture system based on the personalized human body data to provide people with more customized and immersive experiences. In this paper, a rapid and scalable construction method of the wearable inertial measurement unit (IMU) sensor network is proposed to generate personalized wearable solutions for people with different body types. Additionally, a robust self-sensing algorithm based on the IMU sensor network is proposed to reconstruct not only the whole body or limb movements but also the fine-grained muscle deformations. To validate the performance, we evaluate the accuracy and robustness of our method. In the accuracy evaluation, the average measurement error is 3.90mm, less than 1.80% of the test model size (180mm × 150mm × 72mm). In the robustness evaluation, the average measurement error is 6.15mm. Finally, an application on personalized arm motion and deformation capture demonstrates the feasibility and applicability of the proposed self-sensing IMU sensor network.

Unit Commitment (UC) is important for power system operations. With increasing challenges, e.g., growing intermittent renewables and intra-hour net load variability, traditional mathematical optimization could be time-consuming. Machine learning (ML) is a promising alternative. However, directly learning good solutions is difficult in view of the combinatorial nature of UC. This paper synergistically integrates ML within our recent decomposition and coordination method of Surrogate Lagrangian Relaxation to learn “good enough” subproblem solutions of deterministic UC. Compared to original UC, a subproblem is much easier to learn. Nevertheless, predicting good-enough subproblem solutions is still challenging because of the “jumps” of binary decisions and many types of constraints. To overcome these issues, subproblem dimensionality is reduced via aggregating multipliers. Multiplier distributions are novelly specified based on “jumps” for effective learning. Loss functions are innovatively designed to improve prediction qualities. Ordinal Optimization and branch-and-cut are used as backups for unfamiliar cases. Furthermore, online self-learning is seamlessly integrated with offline learning to exploit solutions from daily operations. Results on the IEEE 118-bus system and the Polish 2383-bus system demonstrate that continual learning keeps on improving the subproblem-solving process with near-optimality of the overall solutions maintained. Our method opens a new direction to solve complicated UC.

Tree-based phylogenetic networks provide a powerful model for representing complex data or non-tree-like evolution. Such networks consist of an underlying evolutionary tree called a “support tree” (also known as a “subdivision tree”) together with extra arcs added between the edges of that tree. However, a tree-based network can have exponentially many support trees, and this leads to a variety of computational problems. Recently, Hayamizu established a theory called the structure theorem for rooted binary phylogenetic networks and provided linear-time and linear-delay algorithms for different problems, such as counting, optimization, and enumeration of support trees. However, in practice, it is often more useful to search for both optimal and near-optimal solutions than to calculate only an optimal solution. In the present paper, we thus consider the following problem: Given a tree-based phylogenetic network N where each arc is weighted by its probability, compute the ranking of top-k support trees of N according to likelihood values. We provide a linear-delay (and hence optimal) algorithm for this problem.

The article designs an approach for the detection of MDD using spectral clustering. The raw EEG is pre-processed, and then three quantitative biomarkers: band power (beta, delta, and theta) along with three signal-extracted signals: Detrended Fluctuation Analysis (DFA), Higuchi’s Fractal Dimension (HFD), and Lempel-Ziv Complexity (LZC) are extracted. The data is scrutinized in both inter and intra hemispheric regions. A weighted graph is constructed based on the data that is clustered to obtain the condition of the subject. The results showcase both the efficiency and efficacy of the designed approach. The accuracy achieved for the left hemisphere is 98% and CEP of 2.5%, while 97% accuracy and 3.3% CEP for the right hemisphere. In this case, channel-wise maximum accuracy has been achieved for Fp1 and F8

Line-commutated converter (LCC)-based high-voltage DC (HVDC) systems have been integrated with bulk AC power grids for the interregional transmission of renewable power generation. The nonlinear HVDC model brings additional nonconvexity to the optimal power flow (OPF) of hybrid AC-DC power grids. The convex relaxation of such an OPF model could be a remedy, which has been discussed by little work. In this paper, a convexification method for the OPF of LCC-based AC-DC power grids is proposed using second-order cone programing relaxation. A sufficient condition for the exactness of the proposed convex relaxation is proposed and proven. The properties, exactness and efficiency of the proposed model are verified in three numerical simulations. The simulation results demonstrate that a globally optimal solution of the original OPF can be obtained from the proposed relaxed model with high computational efficiency.

Deep learning is revolutionizing machine learning application domains such as big data analysis, computer vision, and natural language processes. Deep learning is a subset of machine learning that make use of Neural Network(NN) with many processing layers and parameters. This consumes a lot of computing power and memory during the training and testing phases. A common approach to overcoming this computational limitation of deep learning is to use cloud resources. To do this, data to be processed need to be transferred from the source to the cloud. Cloud-based big data processing presents many challenges as the amount of data continues to grow and the data processing requirements differ. Edge computing has emerged to solve these challenges. This article explores how deep learning is used in edge computing and discusses some of the common challenges that have arisen so far.

In this paper, we propose timing offset (TO) and carrier frequency offset (CFO) estimators for orthogonal time frequency space modulation (OTFS). The proposed estimators do not require any additional training overheads as they deploy the same pilot signal that is used for channel estimation. Hence, no additional training overhead is required for synchronization, thanks to the periodic properties of the pilot signal in the delay-time domain. Our proposed TO estimator finds the start of each OTFS block by searching for a periodic sequence in delay and time dimensions. The CFO is then estimated by finding the mean of the angles of a two-dimensional correlation function at the best timing instant. As a novel aspect of these estimators, the multipath diversity of the channel is exploited to achieve a high estimation accuracy. Finally, we show the efficacy of our proposed synchronization techniques through extensive simulations.

This paper elaborates an accurate energy model for Wireless Sensor Device (WSD) based on power consumption measurement. We found that compared to the energy consumption of processing, transmission, listening and reception phases, the data transfer consumes up to 40\% of total WSD energy whereas this aspects is often neglected in usual models.

The circuit design solutions of a new subclass of fourth-order notch filters (NF) are considered, which are implemented on operational amplifiers by sequentially connecting identical second-order links of NF. Equations for their main parameters are given. In the Micro-Cap environment, computer simulation of the amplitude-frequency characteristics was performed, which shows the possibility of independent adjustment of the quality factor of the pole, the frequency of the pole, and the NF transfer coefficient. The proposed circuits differ in the circuitry of the links, as well as in the ways of introducing feedback and options for connecting the input signal source.

Learning graphs represented by M-matrices via an l1-regularized Gaussian maximum-likelihood method is a popular approach, but also one that poses computational challenges for large scale datasets. Recently proposed methods cast this problem as a constrained optimization variant of precision matrix estimation. In this paper, we build on a state-of-the-art sparse precision matrix estimation method and introduce two algorithms that learn M-matrices, that can be subsequently used for the estimation of graph Laplacian matrices. In the first one, we propose an unconstrained method that follows a post processing approach in order to learn an M-matrix, and in the second one, we implement a constrained approach based on sequential quadratic programming. We also demonstrate the effectiveness, accuracy, and performance of both algorithms. Our numerical examples and comparative results with modern open-source packages reveal that the proposed methods can accelerate the learning of graphs by up to 3 orders of magnitude, while accurately retrieving the latent graphical structure of the data. Furthermore, we conduct large scale case studies for the clustering of COVID-19 daily cases and the classification of image datasets to highlight the applicability in real-world scenarios.