Recent network traffic classification methods benefit from machine learning (ML) technology. However, there are many challenges due to use of ML, such as: lack of high-quality annotated datasets, data-drifts and other effects causing aging of datasets and ML models, high volumes of network traffic etc. This paper argues that it is necessary to augment traditional workflows of ML training&deployment and adapt Active Learning concept on network traffic analysis. The paper presents a novel Active Learning Framework (ALF) to address this topic. ALF provides prepared software components that can be used to deploy an active learning loop and maintain an ALF instance that continuously evolves a dataset and ML model automatically. The resulting solution is deployable for IP flow-based analysis of high-speed (100 Gb/s) networks, and also supports research experiments on different strategies and methods for annotation, evaluation, dataset optimization, etc. Finally, the paper lists some research challenges that emerge from the first experiments with ALF in practice.

Modern systems (e.g., assistive technology and self-driving) can place significant demands on the user’s working memory (WM), which can adversely impact performance (i.e., elevated risk of errors) and increase the cognitive load (CL). Robust prediction of CL from EEG remains a challenge due to the small sample problem, noisy recordings, ineffective data representation, and lack of robust models. This paper presents a holistic approach to developing a reliable prediction of CL. We used EEG data recorded following a modified Stenberg WM task in which four levels of CL were defined based on the encoding of 2, 4, 6, and 8 English characters. First, we address the problem of noise and “small sample” by generating large low noise data using eigenspace-based bootstrap sampling and generative adversarial network (GAN). Second, we transform EEG recordings into spatial-spectral images to capture spatial information. Third, we built parameter-optimized CNN models to predict four levels of CL using single-frequency bands (i.e., θ, α, β) and stacked (i.e., all three bands) representations. In our quest to provide interpretable models, we applied Gradientweighted Class Activation Mapping (Grad-CAM) to our models to localize the brain regions responsible for the prediction of CL. Empirical analysis of models trained using θ, α, β, and stacked representation show accuracy of 90%, 89%, 91%, and 94%, respectively. Grad-CAM visualizations showed that the prefrontal, cerebellum, frontal, and parietal areas have the highest contribution to the prediction of CL.

This paper modify the basic Gaussian process regression predictions to get the correct order of magnitude of standard deviation and covariance at some target points different from the data points.

Bifacial photovoltaic modules are able to convert the solar radiation reaching their front and rear-sides, which means that more electricity can be produced using the same array area as monofacial modules with similar ratings. In some locations, the cost per power for such a technology has become cost-competitive with onventional monofacial modules. The so-called effective irradiance and the bifacial gain are useful metrics, respectively, to assess the solar resource and the performance of bifacial arrays. To calculate the effective irradiance, studies previously published employ rear-side irradiance measurements. In turn, to compute the bifacial gain, other works make use of monofacial modules with rating similar to those of the bifacial modules under analysis. In this paper, a straightforward method is presented, allowing the calculation of the effective irradiance without the requirement of measuring the rear-side irradiance; and to calculate the bifacial gain without the need of a monofacial module. The proposed method was experimentally tested using an outdoor platform with a dual-axis tracking system with bifacial modules. The international standard IEC-60891 was taken as a reference for monofacial power calculations, showing coherent results when compared to the calculated front-side bifacial module performance. The method should be useful for the quantification of the effective irradiance and the bifacial gain for PV installations which do not make use of extensive monitoring hardware.

Since hybrid power plants (HPPs) play an intensive role in the energy supply balance of future energy systems, there is today an increased attention on co-located wind-battery HPPs both in industry and academia. However, the profitability of HPPs in sequential electricity markets to overcome initial investment costs has not been yet well examined, especially with respect to  balancing services provision. This article proposes a novel energy management system (EMS) for optimal participation of wind-battery HPPs in two sequential electricity markets, namely in spot market and balancing market. The methodology consists of three optimization models, which allow HPPs to achieve energy arbitrage, to provide balancing services, and to reduce real-time imbalance costs. Furthermore, the profitability of HPPs in future (2030) energy scenario is analyzed based on the new designed and developed EMS taking balancing service into account. The results of the overplanting case show that HPPs operated based on the proposed EMS achieve net present value 3.6 times as high as sole operation in spot market.

The log-sum minimization has been widely employed in signal and image processing. In this paper, in order to analyze its applicable conditions and achievable boundary for compressed sensing scenario, some theoretical derivations of the log-sum function are studied based on the mutual coherence (MC) of measurement matrix. Firstly, the existence condition of the global minimum is demonstrated for the log-sum minimization without noise perturbation. Secondly, for the iterative threshold log-sum (ITL) algorithm, the theoretical performance guarantee of recovering sparse signal disturbed by noise is proposed. Additionally, comprehensive experimental results verify the theoretical analysis for the log-sum one and display the advantages and disadvantages of four sparse estimation algorithms, i.e., the ITL algorithm, ℓ1/2 regularization, ℓ1 regularization and orthogonal matching pursuit (OMP).

The wireless networks of the future are evolving to enable reliable communication for resource-constrained, miniaturized internet of things (IoT) devices, which will place stringent demands on future sixth-generation (6G) mobile networks. These demands include low cost, very low latency, improved spectrum, and power efficiency, higher reliability, and significantly improved data rates. Emphasizing that these devices have limited functionality and may be placed in inaccessible locations, replacing or recharging batteries can be a daunting task, so energy-efficient solutions should be developed to ensure uninterrupted, seamless wireless communication for the power-constrained IoT devices. In this paper, we consider the integration of long-range (LoRa) modulation into backscatter communications (BackCom), and we develop a mathematical framework in order to investigate the error rate performance of the considered system model. In particular, we derive novel exact and approximated closed-form expressions for the symbol error rate (SER), under the assumption of canceled radio-frequency (RF) interference. The obtained analytical results, corroborated by numerical results, confirm the advantages of integrating LoRa into BackCom system as a low-complex technique in order to extend the transmission distance in power-limited backscatter devices.

Since Newton’s Law for Gravitational Fields (G) applies to static reference frames, and is NOT Lorentz invariant like Maxwell’s Equations, we correct it by adding in a Motional Field (S) so F=m/o\ [G + v x S].  Then we unify Gravitation and Electromagnetism by using an Expanded Newton’s Law for Gravitation using Complex Masses, where the complex part is charge associated with the mass.

Pre-trained vision language models (VL) have seen a rise in recent years, achieving state-of-the-art performance on tasks such as visual question answering, image captioning, zero-shot tasks, text-to-image synthesis, and many more. However, these models are large, also requiring large amounts of data, which can hinder application on resource-limited settings. This paper proposes CAPIT (Cross Attention on Pre-trained Image and Text models), a novel architecture built on top of frozen pre-trained unimodal encoders, utilizing transferable knowledge from pre-trainined models through cross-attentional transformers. CAPIT is trained with a simple supervised task that learns to predict the correspondence between image-text pairs. It is then tested on zero-shot image classification. The proposed model contains less parameters and uses less data compared to other methods, while performing fairly well with trade-offs between compute requirements performance, showing promise for future work in resource-limited settings.

Unmanned air vehicles (UAVs) popularity is on the rise as it enables the services like traffic monitoring, emergency communications, deliveries, and surveillance. However, the unauthorized usage of UAVs (a.k.a drone) may violate security and privacy protocols for security-sensitive national and international institutions. The presented challenges require fast, efficient, and precise detection of UAVs irrespective of harsh weather conditions, the presence of different objects, and their size to enable SafeSpace. Recently, there has been significant progress in using the latest deep learning models, but those models have shortcomings in terms of computational complexity, precision, and non-scalability. To overcome these limitations, we propose a precise and efficient multiscale and multifeature UAV detection network for SafeSpace, i.e., \textit{MultiFeatureNet} (\textit{MFNet}), an improved version of the popular object detection algorithm YOLOv5s. In \textit{MFNet}, we perform multiple changes in the backbone and neck of the YOLOv5s network to focus on the various small and ignored features required for accurate and fast UAV detection. To further improve the accuracy and focus on the specific situation and multiscale UAVs, we classify the \textit{MFNet} into small (S), medium (M), and large (L): these are the combinations of various size filters in the convolution and the bottleneckCSP layers, reside in the backbone and neck of the architecture. This classification helps to overcome the computational cost by training the model on a specific feature map rather than all the features. The results show significant performance gain even for unseen feature maps with minimal loss in accuracy. Results show a significant reduction in training parameters, inference, and increased pattern in FPS and GFLOPs for \textit{MFNet} compared to YOLOv5s. \textit{MFNet-M} performance evaluation in terms of precision, recall, mean average-precision (mAP), and IOU increased around 1.8\%, 2.2\%, 0.9\%, 1.7\% compared to YOLOv5s. Furthermore, \textit{MFNet-M} achieves the best performance with 96.8\% precision, 88.4\% recall, 95.9\% mAP, and 51.1\% IoU for UAV detection. The dataset and code are available as an open source: github.com/ZeeshanKaleem/MultiFeatureNet.

We present a method to implement signal averaging using wavelet transforms. Signal averaging plays an essential role in reducing noise from physical experiments, but is currently performed outside wavelet transforms and removes the information of individual data points or experimental scans. We introduce the conditions that need to be satisfied to implement signal averaging using wavelet transforms, showing that only the Haar wavelet can be used to obtain signal-averaged information. Our method identifies the wavelet component that contains the signal-averaged value and develops a procedure to recover it. The method is tested on model data using varied amounts of Gaussian and Poisson noise at different data lengths as well as experimental data from the Electron Spin Resonance (ESR) spectroscopy. While the Haar wavelet is successful in obtaining the mean, other wavelets yield weighted averages. The incorporation of this method enables signal averaging and analysis to be carried out simultaneously in the wavelet domain, expanding the application of wavelet transforms and improving wavelet analysis. The code is available via GitHub and denoising.cornell.edu, as well as the corresponding author’s group website (signalsciencelab.com).

This work investigates the effects of visual evoked potentials on P300-based brain-computer interfaces and proposes a regularized version of the xDAWN algorithm to improve these systems.

This paper describes a new optical spectroscopy instrument achieving lock-in detection in all its spectral output channels in a parallel and simultaneous fashion.  It is based on two identical compact array spectrometers working in phase quadrature thanks to optical modulators synchronized with an external reference signal. In the application example, the optical lock-in spectrometer was used to perform spectral irradiance measurements with commercial LED lamps exhibiting temporal light modulation at different frequencies. A remote sensing optical device was designed to measure from a distance the temporal luminous waveform of a specific LED lamp and send it to the reference input of the optical lock-in spectrometer. When the lamps were operating together on top of a large continuous luminous background, the optical lock-in spectrometer successfully retrieved the respective spectral irradiance distributions and color parameters of each lamp, allowing their characterization to be performed independently of the other light sources illuminating the sensor during the measurements. © 2023 IEEE.  Personal use of this material is permitted.  Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works

Cardiovascular diseases are the leading causes of death, and blood pressure (BP) monitoring is essential for prevention, diagnosis, assessment, and treatment. Photoplethysmography (PPG) is a low-cost opto-electronic technique for BP measurement that allows the acquisition of a modulated light signal highly correlated with BP. There are several reports of methods to estimate BP from PPG with impressive results; in this study, we demonstrate that the previous results are excessively optimistic because of their train/test split configuration. To manage this limitation, we considered intra- and inter-subject data arrangements and demonstrated how they affect the results of feature-based BP estimation algorithms (i.e., XGBoost, LightGBM, and CatBoost) and signal-based algorithms (i.e., Residual U-Net, ResNet-18, and ResNet-LSTM). Inter-subject configuration performance is inferior to intra-subject configuration performance, regardless of the model. We also showed that, using only demographic attributes (i.e., age, sex, weight, and subject index number), a regression model achieved results comparable to those obtained in an intra-subject scenario. Although limited to a public clinical database, our findings suggest that algorithms that use an intra-subject setting without a calibration strategy may be learning to identify patients and not predict BP.

The rise of adaptive stimulation approaches has shown great therapeutic promise in the growing field of neuromodulation. The discovery and growth of these novel adaptive stimulation paradigms has been largely concentrated around several implantable devices with research application programming interfaces (APIs) that allow for custom applications to be created for clinical neuromodulation studies. However, the sunsetting of devices and ongoing development of new platforms is leading to an increased fragmentation in the research environment- resulting in the reinvention of system features and the inability to leverage previous development efforts for future studies. The Open Mind Neuromodulation Interface (OMNI) is a previously proposed solution to address the weaknesses of the DLL-driven API approach of past neuromodulation research by utilizing an alternative gRPC-enabled microservice framework. Here, we introduce OMNI-BIC, an implementation of the OMNI framework to the CorTec Brain Interchange system. This paper describes the design and implementation of the OMNI-BIC software tools and demonstrates the framework’s capabilities for implementing customized neuromodulation therapies for clinical investigations. Through the development and deployment of the OMNI-BIC system, we hope to improve future clinical studies with the Brain Interchange system and aid in continuing the growth and momentum of the exciting field of adaptive neuromodulation.

We describe a portable 32-antenna coherent passive radar system. The system was developed with space domain awareness in mind, but can be used for aircraft and nearer objects. We describe the physical and signal processing architecture and the results of a system test using VHF-band broadcast transmitters as illuminators. Using a combination of phased-array beamforming and delay/Doppler tracking, the flight path of a small training aircraft was tracked with high accuracy in excellent agreement with public flightpath data.

The housing crisis in Ireland has rapidly grown in recent years. To make a more significant profit, many landlords are no longer renting out their houses under long-term tenancies but under short-term tenancies. The shift from long-term to short-term rentals has harmed the supply of private housing rentals. Regulating rentals in Rent Pressure Zones with the highest and rising rents is becoming a tricky issue. In this paper, we develop a breach identifier to check short-term rentals located in Rent Pressure Zones with potential breaches only using publicly available data from Airbnb (an online marketplace focused on short-term home-stays). First, we use a Residual Neural Network to filter out outdoor landscape photos that negatively impact identifying whether an owner has multiple rentals in a Rent Pressure Zone. Second, a Siamese Neural Network is used to compare the similarity of indoor photos to determine if multiple rental posts correspond to the same residence. Next, we use the Haversine algorithm to locate short-term rentals within a circle centered on the coordinate of a permit. Short-term rentals with a permit will not be restricted. Finally, we improve the occupancy estimation model combined with sentiment analysis, which may provide higher accuracy. Because Airbnb does not disclose accurate house coordinates and occupancy data, it is impossible to verify the accuracy of our breach identifier. The accuracy of the occupancy estimator cannot be verified either. It only provides an estimate within a reasonable range. Users should be skeptical of short-term rentals that are flagged as possible breaches.

Background: Amongst many suggestions toward fusion power ignition, particle acceleration related model has a advantage of reaching high temperature performance. Objective: Suggesting a plasma enclosure model by applying particle acceleration could be tried. Method: By given equation from unification of energy, application of heat and work transformation gives model to ignition of fusion power by accelerating fusion fuel. Conclusion: By accelerating multiple inputs of fuel for fusion in an initiation of fusion power, higher temperature that meets requirement for nuclear fusion could be established.