The design of a fully integrated adaptive modified complementary metal-oxide-semiconductor (CMOS) synapse circuit is presented. By using multiple-gated transistor configuration in the modified CMOS synapse an additional branch provide control where the synaptic output current time-constant is tuned. The effect of changing the multiple-gated transistor bias voltage from 0.25 to 0.45 V tunes the spiking output current exponential time-constant range by 200 ms as shown in simulation results. Moreover, a fully-integrated adaptive quadratic integrate-and-fire (QIF) CMOS neuron circuit is presented as well. A differential pair with variable capacitor integrator and a tunable schmitt trigger threshold detector circuit are integrated in the CMOS neuron that can be tuned varying its spiking frequency. The proposed adaptive quadratic integrate-and-fire (AQIF) neuron has the ability to adjust the spiking frequency without changing the input current. The simulation results show the proposed CMOS neuron circuit spiking frequency can be tuned from 58.4 to 312.5 Hz and its spiking period from 17.1 to 3.2 ms with tuning the bias voltage of variable capacitor integrator. Having a peak voltage Vpeak=0.95 V, a reset voltage Vreset=-0.75 V and a voltage threshold of 0.35 V with a membrane potential range of 1.5 V. The proposed CMOS neuron circuit is designed in 130 nm process with a supply voltage of 1.8 V and a total power dissipation of 1.8 mW.

Financial time series forecasting faces significant challenges due to data scarcity, high volatility, and inherent nonlinearities. Complex deep learning models, such as transformers, typically require extensive datasets and computational resources, making them prone to overfitting in financial contexts where datasets are limited. To address this, we propose Augmented Reverse Training LSTM (ART-LSTM), a novel data augmentation strategy for time series forecasting using a straightforward unidirectional LSTM architecture. ART-LSTM leverages both forward and reversed sequences during training, effectively doubling the available training data without increasing architectural complexity. Our approach maintains computational simplicity while enhancing model robustness and generalisation. Empirical evaluations on challenging datasets, including daily S&P 500 index prices and USD/EUR exchange rates, demonstrate that ART-LSTM consistently outperforms traditional statistical methods (ARIMA), standard recurrent neural networks (RNN, GRU, and LSTM), and multi-layer perceptrons (MLPs), achieving substantial reductions in Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Overall, ART-LSTM provides a practical and data-efficient solution for financial forecasting tasks characterised by limited data availability and volatile dynamics.

The increasing size of machine learning models and the datasets used for training has resulted in significantly higher computational demands. Modern large language models, in particular, consume vast amounts of energy during training, raising serious environmental concerns. In this paper, we propose a low-energy classifier, the Undersampled Random Forest (URF), designed specifically for imbalanced data. The URF method leverages undersampling in combination with a parallelizable random forest architecture. This approach reduces the size of the training data and accelerates computation, leading to more energy-efficient training processes. Moreover, the proposed method consistently achieves high recall rates on minority class instances, making it both effective and environmentally conscious.

This work presents a new controller for gridconnected PV/Battery systems that combines a bidirectional battery controller with a voltage source converter (VSC) to solve problems caused by power fluctuations on the DC-link. The hybrid VSC controller is developed to maintain a continuous power supply on the DC-link. It does this by responding to changes in environmental factors and load fluctuations in the PV-generated output, which in turn triggers dynamic adjustments to the inverter pulse signals. To increase system stability, the bidirectional battery controller orchestrates the charge and discharge operations in collaboration with the hybrid VSC, optimizing power delivery to the grid. Through a detailed performance study of the proposed controller, faults on both the DC and AC sides of the system are analyzed and efficiently addressed. The controller proved important in enhancing power quality and decreasing harmonic distribution. Furthermore, the hybrid controller's performance is assessed under several one-day environmental situations. Ultimately, the results of simulation done in MATLAB/Simulink corroborate the usefulness of the suggested controller in providing optimal performance in the described grid-connected PV/Battery system.

We present an automatic signal modulation classification model using combinatorial deep learning technique. Our proposed deep learning model increase accuracy for low Signal-to-Noise Ratio (SNR) and maintain a high classification accuracy for high SNR signals. Using a hybrid deep learning model combining both ConvLSTM with Transformer-block neural networks, the proposed modulation classifier architecture can learn the signal for both low and high SNR and get better accuracy for signals with high noise. The proposed deep learning modulation classification technique achieves improved classification accuracy of 66% for low SNR signals and 93.5% at high SNR showing that our model is robust under noisy signal modulation. Thus, getting better accuracy in lower SNR signals without sacrifice accuracy for higher SNR signals. An adaptive weighted focal loss function is proposed as an optimized loss function for efficient classification which can be used to control the outliers within a class imbalance and avoid underflow issues. Our deep learning radio modulation classification model works using raw signal without the need of denoising the noisy signal.

The study addresses the conformity of actual electricity consumption to the calculated value in electric distribution networks in which municipal consumers predominate in several cities of the Chelyabinsk region. To study the conformity between the specific electrical load established by regulatory documents and the actual value per apartment according to power consumption data in several cities of the Chelyabinsk region, the average annual power consumption by municipal consumers with a specific number of apartments was analyzed over a period of 2021–2022. The correspondence analysis of the average annual electricity consumption by municipal consumers in the studied facilities was carried out using the conventional method for calculating the electrical load over the given period following the guidelines outlined in SP 256.1325800.2016. The discrepancy between the actual electrical load on the apartment and its normative value established by the acting normative documents ranged from minus 48 to 300% with respect to electricity consumption. For the considered 16 objects located in the cities of the Chelyabinsk region, the discrepancy between the actual electrical load and the established normative values was compared. For 6 apartments, this discrepancy ranged from minus 58 to 155%. To improve the accuracy of forecasting electricity consumption and calculating electrical loads in electric distribution networks with a predominance of municipal consumers, methods using a new factor were recommended. This factor involves a generalized uncertainty coefficient Ai, whose values are determined for the considered period. When using the developed methods, relative deviations in the forecast calculations are less than or equal to 10%.

A neural network-based parking system with real-time license plate detection and vacant space detection using hyper parameter optimization is presented. When number of epochs increased from 30, 50 to 80 and learning rate tuned to 0.001, the validation loss improved to 0.017 and training object loss improved to 0.040. The model mean average precision mAP_0.5 is improved to 0.988 and the precision is improved to 99%. The proposed neural network-based parking system also uses a regularization technique for effective predictive modeling. The proposed modified lasso ridge elastic (LRE) regularization technique provides a 5.21 root mean square error (RMSE) and an R-square of 0.71 with a 4.22 mean absolute error (MAE) indicative of higher accuracy performance compared to other regularization regression models. The advantage of the proposed modified LRE is that it enables effective regularization via modified penalty with the feature selection characteristics of both lasso and ridge.

Technology plays a pivotal role in modern university curriculum. Given the recent advances in machine learning, it is particularly important to leverage the tools provided by the latest advances in computing. In this paper, we investigate the effects of including Python-based computer algebra system SymPy in university calculus courses. The advantages of SymPy include open-source, introduction to programming, and simplicity. Critically, comparison of the grades pre and post adoption of SymPy in calculus courses indicates a statistically significant improvement. Although further studies - based on a larger cohort - about the effectiveness of open-access computer algebra system are desirable, the initial findings suggest that SymPy provides a viable alternative to commercially available computer algebra system.

In this paper, the design of an RF LNA with simultaneous noise-cancellation and distortion-cancellation is presented. The proposed LNA simulation results show a S21 gain of 15.4 dB and total noise Figure cancellation of 0.45dB with a circuit noise Figure NF of 1.6 dB and S12 reverse isolation of 38.2 dB at frequency 2.4 GHz. The simulation results show a 12 dBc of IM3 distortion-cancellation and 6 dB IIP3 improvement with power consumption of 9.5 mA. Proposed RF LNA simultaneous cancellation of noise and distortion improves circuit noise Figure and linearity and in turn improves the overall wireless RF system dynamic range.

This paper presents a Lora-enabled GPU-based CubeSat Neural-Network Real-Time Object Detection with hyperparameter optimization is presented. When number of epochs increased from 10 to 50 and learning rate tuned to 0.00104, the validation loss improved to 0.018 and training object loss improved to 0.042. Model mean average precision mAP_0.5 is improved to 0.986 and the precision is improved to 98.9%. Epoch and learning rate are traded-off to optimize model accuracy performance. The Lora-enabled CubeSat onboard transceiver provides long range onboard sensors readings providing spaced-based IoT application capability.

Recent advancements in technologies enabled the development of smart cities to be more effective and possible. Smart cities depend on intelligent systems, artificial intelligence, the internet of things, control system, and many more advanced technologies. Sustainability challenges and problems worldwide, with smart and sustainability concepts, reflect almost mutual goals. It includes improving and providing the essential life services for all people efficiently while depending on sustainable, clean, and renewable energy with considerations of different economic, educational, health, social and environmental aspects in the city. In this research, a cost analysis process has been implemented to ease the implementation and resource utilization of smart and sustainable cities in Africa. The challenges and difficulties of those implementations are summarized.

In this paper, we apply a fusion machine learning method to construct an automatic intrusion detection system. Concretely, we employ the orthogonal variance decomposition technique to identify the relevant features in network traffic data. The selected features are used to build a deep neural network for intrusion detection. The proposed algorithm achieves 100% detection accuracy in identifying DDoS attacks. The test results indicate a great potential of the proposed method.

This paper presents the design of a fully-integrated RF low noise amplifier with active inductor linearizer. The proposed circuit achieves 16 dBc of third order intermodulation IM3 distortion cancellation at 1.9 GHz with 8 dB third-order intercept point IP3 improvement. The proposed circuit utilize an active inductor linearizer to improve linearity of the circuit and offering tunability. The simulation results show an overall proposed circuit peak gain is 14.5 dB and the minimum noise Figure is 0.75 dB at 1.9 GHz frequency with power consumption of 7.4 mA.

This paper presents the design of a fully-integrated tunable Q-enhanced LNA resonator filter designed to tune the circuit center frequency and quality factor Q. The proposed circuit achieves a 600 MHz 3dB bandwidth tunable center frequency at 2.4 GHz with a 5.5 dB Quality Factor Q tuning range. The proposed circuit utilize a distortion transistor compensator to improve linearity of the circuit. The results show an 18 dBc of third order intermodulation IM3 cancellation. The overall proposed circuit peak gain is 16.5 dB and the minimum NF is 0.94 dB at 2.4 GHz frequency with power consumption of 5.2 mA.