Random Search Research Articles

Multi-objective optimization in high-dimensional patent layout using deep Bayesian network and hybrid algorithm

This study introduces a novel multi-objective optimization framework (DBNO) that integrates deep Bayesian networks with a hybrid algorithm combining random search and innovation diffusion to address high-dimensional patent layout optimization challenges. The framework was developed in response to the increasing complexity of patent layout decisions, where traditional single-objective optimization methods prove inadequate for simultaneously addressing multiple conflicting objectives such as profit, risk, and sustainability. To evaluate the framework's effectiveness, we conducted comprehensive experiments comparing DBNO against established algorithms including genetic algorithm (GA), particle swarm optimization (PSO), and traditional Bayesian optimization methods. Performance metrics encompassed convergence speed, computational efficiency, optimization stability, and solution quality across multiple objectives. The results demonstrate that DBNO consistently outperforms benchmark algorithms, particularly in optimizing sustainability objectives. Notably, DBNO exhibited superior stability and higher success rates in the optimization process compared to GA and PSO, highlighting its robustness in handling complex high-dimensional optimization problems. Furthermore, the integration of innovation diffusion mechanisms significantly enhanced both the efficiency and accuracy of the optimization process. The primary contribution of this research lies in the novel combination of deep Bayesian networks with ensemble random search techniques, resulting in a powerful multi-objective optimization framework. This approach provides an effective solution for high-dimensional patent layout problems while offering new perspectives for patent strategic decision-making. The findings advance the field of multi-objective optimization and establish a foundation for future research in patent portfolio optimization.

Ultra-Short-Term Load Forecasting for Extreme Scenarios Based on DBSCAN-RSBO-BiGRU-KNN-Attention with Fine-Tuning Strategy

Extreme scenarios involving abnormal load fluctuations pose serious challenges to the safe and stable operation of power systems. To address these challenges, an ultra-short-term load forecasting model is proposed, specifically designed for extreme conditions. The model combines density-based spatial clustering of applications with noise (DBSCAN), random search Bayesian optimization (RSBO), bidirectional gated recurrent units (BiGRUs), k-nearest neighbor (KNN), and an attention mechanism, enhanced by a fine-tuning strategy to improve forecasting accuracy. Firstly, the original load data are reconstructed weekly, and extreme scenarios are identified using the DBSCAN. Secondly, the RSBO is employed to optimize model parameters within the high-dimensional search space. To further refine performance, the final fully connected layer is fine-tuned to adapt to extreme conditions. Finally, case studies demonstrate that the proposed approach reduces the root mean square error (RMSE) by 12.37% and the mean absolute error (MAE) by 6.73% compared to benchmark models, achieving superior accuracy under all tested extreme scenarios.

Применение модификации метода, имитирующего поведение стаи мотыльков, для решения задачи оптимального программного управления мобильным роботом

<p>The article proposes a modification of the metaheuristic optimization method simulating the behavior of a swarm of moths, which belongs to the group of bio-inspired methods. A step-by-step algorithm is formed and the efficiency of the modified method is studied on a generally accepted set of test functions of many variables with a complex structure of level surfaces, on the problem of optimal open loop control with a known exact solution, as well as on the problem of determining the parameters of the tension/compression spring with constraints such as inequalities. The advantage of the modified method over the original version is shown. It is demonstrated that the method allows finding solutions of sufficiently good quality in a time acceptable from a practical point of view. A solution to an applied problem of finding the optimal open loop control of a mobile robot on a plane in the presence of obstacles is given. The goal of the control is to achieve a given end point while minimizing the elapsed time and fulfilling the condition of enveloping forbidden areas. To find the control law as a function of time, a piecewise constant approximation was used, which allows reducing the problem to finding a finite number of unknown parameters. The solutions of the terminal problem of speed of response with different structure and parameters of the composite quality functional are considered using the sequential application of the developed modified method simulating the behavior of a swarm of moths, the random search method with sequential reduction of the study area and the path-relinking method. The results of comparison with known solutions are presented.</p>

First-passage times for generalized heterogeneous telegrapher's processes

We consider two different fractional generalizations of the heterogeneous telegrapher's process with and without stochastic resetting. Both governing fractional heterogeneous telegrapher's equations can be obtained from the corresponding standard heterogeneous telegrapher's equations by using the subordination approach. The first-passage time problems are solved analytically for both models by finding the survival probabilities, the first-passage time densities, and the mean first-passage times. We showed that for both cases there are optimal resetting rates for which the mean first-passage times are minimal. The present work carries implications toward our understanding of anomalous diffusion and random search in heterogeneous media. Published by the American Physical Society 2025

A stacked ensemble machine learning model for the prediction of pentavalent 3 vaccination dropout in East Africa

IntroductionVaccination is critical for reducing childhood mortality, yet completion rates for the third dose of the pentavalent vaccine (Penta 3) in East Africa remain inadequate. This study aims to predict Penta 3 vaccination dropout using a stacking ensemble machine learning model with Demographic and Health Survey (DHS) data. The objective is to identify predictors of dropout and enhance intervention strategies.MethodsThe study utilized seven base machine learning algorithms to create a stacked ensemble model with three meta-learners: Random Forest (RF), Generalized Linear Model (GLM), and Extreme Gradient Boosting (XGBoost). The H2O package facilitated the development of base learners and the stacking of super learners. Feature selection (FS) and comparisons were performed using the LASSO and Boruta algorithms. The selected features were one-hot encoded, and ordinal encoding was applied where appropriate. Hyperparameter optimization (HPO) and comparisons were conducted using grid search and random search. Model performance was assessed using five key metrics, including accuracy and the area under the curve (AUC). SHAP (Shapley Additive Explanations) values were used to interpret the model outputs and identify influential predictors. The experimental design was employed to present the results.ResultsFour experiments were conducted to evaluate feature selection and HPO methods. All stacked ensemble models outperformed individual learners, with the XGBoost meta-learner optimized with grid search and LASSO FS achieving the highest performance: 93.9% accuracy and 99.4% AUC. While RF and GLM meta-learners were also evaluated, they were outperformed by the XGBoost meta-learner. SHAP analysis revealed key features influencing Penta 3 dropout, including the place of delivery, decision-making autonomy, the mother's level of earning, and healthcare access. Home delivery increased the risk of dropout, while postnatal care by midwives and health insurance coverage lowered dropout likelihood.Conclusion and recommendationThis study provides insights into the factors influencing Penta 3 vaccination dropout in East Africa. To reduce dropout rates, interventions should focus on enhancing maternal livelihood opportunities, improving healthcare access in rural areas, and promoting institutional deliveries.

Optimization Algorithm for Cutting Masonry with a Robotic Saw

The contribution of this study is in the novel application of the bin packing algorithm that is used to optimize the robotic bricklaying process with the aim of minimizing the wearing of a robotic saw used for splitting brick blocks so as to minimize brick consumption. To optimize the cutting of masonry blocks with a robotic saw, a new bin packing algorithm has been developed to enhance the design of a digital cutting plan. The algorithm is based on the principle of random search for all combinations of cutting execution with respect to the maximum number of objects (cuts) found in one container (masonry block). The new bin packing algorithm (NBPA) minimizes the number of total masonry blocks (containers) and the number of cuts made with a robotic saw, thus reducing the cutting length. The algorithm can converge to a solution rather quickly and reliably to identify optimal variants of a digital plan designed for a robotic saw to be used in different object assemblies. This article describes the optimization algorithm, including step-by-step calculations, and provides a practical example and a comparison of the results with earlier algorithms. The concept of the robotic saw is also presented in detail, including a description of a prototype. The simulation of the performance on 20 different sets of elements showed that NBPA has a similar use of space compared to the First-Fit Decreasing algorithm (FFD). Multicriteria analysis demonstrated that when the weighting criterion for saw wear was 40% of all the criteria, the use of NBPA was approximately 3.5 times more effective than FFD. The application of the new methodology to a robotic bricklaying process has the potential to reduce the wear of robotic saw, to increase the speed of the construction process and to reduce the generation of construction and demolition waste (CDW).

Predicting determinants of unimproved water supply in Ethiopia using machine learning analysis of EDHS-2019 data

Over 2 billion people worldwide are impacted by the global dilemma of access to clean and safe drinking water. The problem is most acute in low-income nations, where many people still use unimproved water sources such as exposed wells and surface water. Public health systems are heavily burdened by these sources since they are closely associated with the spread of waterborne illnesses. As a result, there are still many people who suffer from water-related health problems, especially in underdeveloped nations where access to healthcare is limited and sanitation is often inadequate. However, the conventional analytical techniques employed in these investigations frequently fall short of capturing the intricate relationships among many variables, which could restrict the capacity to forecast future patterns. This study aimed to provide more accurate predictions and data-driven insights that can inform policy-making, resource allocation, and interventions to address Ethiopia’s water crisis. The Ethiopia Demographic and Health Survey (EDHS-2019), which offers thorough data on socioeconomic, demographic, and water access determinants, was the data source for this study. The following six machine-learning models were used: k-nearest Neighbors, Random Forest, Support Vector Machines, Gradient Boosting Machines, and Artificial Neural Networks. To enhance model performance and prevent overfitting, Hyperparameter adjustment was accomplished via random search and 7-fold cross-validation. The model’s performance was evaluated using the standard classification metrics (accuracy, precision, recall, F1-score, and AUC). To examine the significance of features in tree-based models, permutation importance and SHAP values were utilized. In important measures such as AUC (0.8915), F1 Score (0.919), sensitivity (0.879), and specificity (0.967), the Random Forest model fared better than the other models. “Community-level poverty” was the most important predictor, followed by “household wealth index” and “age of household head,” according to feature importance analysis. Geographic differences in access to better water sources were found through spatial analysis, with rural areas being the most impacted. Using machine-learning algorithms, specifically Random Forest, has yielded significant insights into the factors influencing Ethiopia’s unimproved water supply. The results highlight the necessity of focused interventions in areas with high rates of poverty and insufficient infrastructure. These data-driven insights can help decision-makers better solve Ethiopia’s water crisis.

Adaptive Gait Control for Quadruped Robots on Varied Slopes via ARS Algorithm

Ensuring stable walking for quadruped robots on unknown slopes is a critical challenge in robotic navigation. This study introduces a novel gait planning algorithm that leverages data from an Inertial Measurement Unit (IMU) for terrain slope estimation, offering a cost-effective alternative to visual sensors. The proposed approach integrates a trot gait with an elliptical foot trajectory, enabling efficient movement across varied slopes. Using the Augmented Random Search (ARS) algorithm, we fine-tune the elliptical trajectory parameters to achieve precise and adaptive foot placements. Additionally, the robot dynamically adjusts its posture in real time to maintain stability by aligning with desired joint angles during slope traversal. Simulation results validate the effectiveness of the proposed algorithm, demonstrating its ability to ensure stable and adaptive locomotion on slopes of up to 11 degrees. This work highlights the feasibility of using low-cost hardware and advanced algorithms to address complex terrain navigation challenges.

Optimizing deep learning models from multi-objective perspective via Bayesian optimization

Optimizing hyperparameters is crucial for enhancing the performance of deep learning (DL) models. The process of configuring optimal hyperparameters, known as hyperparameter tuning, can be performed using various methods. Traditional approaches like grid search and random search have significant limitations. In contrast, Bayesian optimization (BO) utilizes a surrogate model and an acquisition function to intelligently navigate the hyperparameter space, aiming to provide deeper insights into performance disparities between naïve and advanced methods. This study evaluates BO's efficacy compared to baseline methods such as random search, manual search, and grid search across multiple DL architectures, including multi-layer perceptron (MLP), convolutional neural network (CNN), and LeNet, applied to the Modified National Institute of Standards and Technology (MNIST) and CIFAR-10 datasets. The findings indicate that BO, employing the tree-structured parzen estimator (TPE) search method and expected improvement (EI) acquisition function, surpasses alternative methods in intricate DL architectures such as LeNet and CNN. However, grid search shows superior performance in smaller DL architectures like MLP. This study also adopts a multi-objective (MO) perspective, balancing conflicting performance objectives such as accuracy, F1 score, and model size (parameter count). This MO assessment offers a comprehensive understanding of how these performance metrics interact and influence each other, leading to more informed hyperparameter tuning decisions.

Comparative study of Neural Architecture Search methods: Random Search, RNN + Reinforcement Learning, and Evolutionary Algorithms on CIFAR-10

Comparative study of Neural Architecture Search methods: Random Search, RNN + Reinforcement Learning, and Evolutionary Algorithms on CIFAR-10

Reinforcement learning using neural networks in estimating an optimal dynamic treatment regime in patients with sepsis.

Reinforcement learning using neural networks in estimating an optimal dynamic treatment regime in patients with sepsis.

Deep Learning-Enhanced Motor Training: A Hybrid VR and Exoskeleton System for Cognitive–Motor Rehabilitation

This research explored the integration of the real-time machine learning classification of motor imagery data with a brain–machine interface, leveraging prefabricated exoskeletons and an EEG headset integrated with virtual reality (VR). By combining these technologies, the study aimed to develop practical and scalable therapeutic applications for rehabilitation and daily motor training. The project showcased an optimized system designed to assess and train cognitive–motor functions in elderly individuals. Key innovations included a motor imagery EEG acquisition protocol for data classification and a machine learning framework leveraging deep learning with a wavelet packet transform for feature extraction. Comparative analyses were conducted with traditional models such as Support Vector Machines (SVMs), Convolutional Neural Networks (CNNs), and Long Short-Term Memory (LSTM) networks. The performance was further enhanced through a random hyperparameter search, optimizing feature extraction and learning parameters to achieve high classification accuracy (89.23%). A novel VR fishing game was developed to dynamically respond to EEG outputs, enabling the performance of interactive motor imagery tasks in coordination with upper limb exoskeleton arms. While clinical testing is ongoing, the system demonstrates potential for increasing ERD/ERS polarization rates in alpha and beta waves among elderly users after several weeks of training. This integrated approach offers a tangible step forward in creating effective, user-friendly solutions for motor function rehabilitation.

An optimization model for the marine current turbine installation problem addressing parameter ambiguity

To enhance the design and layout of marine farms, while addressing the ambiguity of environmental parameters, this study proposes a non-parametric, scenario-based robust bi-level optimization model for the marine current turbine installation problem (BO-MCTIP). In this model, the upper level determines the number and locations of marine current turbines (MCTs), while the lower level determines how to connect the installed MCTs. First, all scenarios in the model are solved using a novel greedy heuristic algorithm (GHA), which minimizes the cost per watt by considering wake effects and submarine power cable connection costs. GHA outperforms the complete search and random search algorithms. Then, using the solutions obtained by GHA in the BO-MCTIP, the min–max relative regret decision rule is used to identify a robust solution. The result from a test case in Cook Inlet, Alaska, demonstrates that the robust solution performs effectively across all scenarios considered in the proposed model.

Optimized machine learning-based enhanced modeling of pile bearing capacity in layered soils using random and grid search techniques

The bearing capacity of a pile is a critical factor in geotechnical design, necessitating extensive testing procedures that often increase both the time and cost of earthwork. Consequently, there is a growing demand for efficient and reliable methods to determine pile bearing capacity. This study aims to propose optimized machine learning based models through the application of Random Search (RS) and Grid Search (GS) optimization techniques for the prediction of pile-bearing capacity in layered soils. For this purpose, an extensive dataset is sourced from literature, and various machine learning algorithms including Random Forest (RF), Support Vector Machine (SVM), and XGBoost are investigated. Through a systematic modeling approach, multiple models are generated, and the performance of machine learning algorithms is refined using RS and GS cross validation (CV) using a customized code in Python. Optimized models are further assessed based on comprehensive evaluation criteria using key statistical performance indices. The results demonstrate that both RS and GS-tuned machine learning models achieve high accuracy, with R2 values exceeding 0.9 and a low error index score across testing and training datasets. Notably, GS exhibits slightly superior statistical performance compared to RS. Furthermore, the tuned models with RS and GS showcase high performance on the validation dataset, with GS consistently outperforming RS. XGBoost emerges as the top performer among the machine learning models, followed by RF and SVM, highlighting the efficacy of tree-based algorithms in capturing the geotechnical variability inherent within pile bearing data. The proposed models offer valuable insights for predicting the preliminary evaluation of pile bearing capacity, facilitating swift and cost-effective geotechnical characterization within an acceptable error margin. This study introduces advancements in predictive modeling for geotechnical engineering, highlighting the transformative potential of optimization methodologies to enhance the machine learning models used for decision-making processes in civil engineering applications.

Comparative Analysis of Reinforcement Learning Algorithms for Finding Reaction Pathways: Insights from a Large Benchmark Data Set.

The identification of kinetically feasible reaction pathways that connect a reactant to its product, including numerous intermediates and transition states, is crucial for predicting chemical reactions and elucidating reaction mechanisms. However, as molecular systems become increasingly complex or larger, the number of local minimum structures and transition states grows, which makes this task challenging, even with advanced computational approaches. We introduced a reinforcement learning algorithm to efficiently identify a kinetically feasible reaction pathway between a given local minimum structure for the reactant and a given one for the product, starting from the reactant. The performance of the algorithm was validated using a benchmark data set of large-scale chemical reaction path networks. Several search policies were proposed, using metrics based on energetic or structural similarity to the product's goal structure, for each local minimum structure candidate found during the search. The performances of baseline greedy, random, and uniform search policies varied substantially depending on the system. In contrast, exploration-exploitation balanced policies such as Thompson sampling, probability of improvement, and expected improvement consistently demonstrated stable and high performance. Furthermore, we characterized the search mechanisms that depend on different policies in detail. This study also addressed potential avenues for further research, such as hierarchical reinforcement learning and multiobjective optimization, which could deepen the problem setting explored in this study.

WACSO: Wolf Crow Search Optimizer for Convolutional Neural Network Hyperparameter Optimization

Convolutional Neural Networks (CNNs) experience performance and training efficiency changes according to the selection of correct hyperparameters. The research presents WACSO which combines Crow Search Optimization with Grey Wolf Optimizer to improve Convolutional Neural Networks hyperparameter selection through a hybrid metaheuristic algorithm. The hybrid algorithm WACSO uses exploration parts from CSO together with GWO exploitation mechanics to obtain optimized performance. WACSO reaches higher classification accuracy than traditional optimization algorithms when performing tests on the MNIST and CIFAR-10 datasets along with Random Search and particle swarm optimization and genetic algorithms and standalone CSO and standalone GWO. The best classification results reached 98.9% accuracy levels on MNIST along with 91.5% accuracy levels on CIFAR-10. The final outcomes of this system depend on the combination of model structure along with dataset challenges and available computational power. The investigation demonstrates that mixing algorithms drawn from nature can lead to successful CNN hyperparameter optimization. The promising outcomes of WACSO depend on multiple variables including computation expenses and sensitive parameter adjustments and universal result adaptability between different datasets and network setups. Research into WACSO should expand to involve longer evaluations across multiple datasets and various models to confirm widespread usage.

Estimation of Gross Primary Productivity Using Performance-Optimized Machine Learning Methods for the Forest Ecosystems in China

Gross primary productivity (GPP) quantifies the rate at which plants convert atmospheric carbon dioxide into organic matter through photosynthesis, playing a vital role in the terrestrial carbon cycle. Machine learning (ML) techniques excel in handling spatiotemporally complex data, facilitating accurate spatial-scale inversion of forest GPP by integrating limited ground flux measurements with Remote Sensing (RS) observations. Enhancing ML algorithm performance for precise GPP estimation is a key research focus. This study introduces the Random Grid Search Algorithm (RGSA) for hyperparameters tuning to improve Random Forest (RF) and eXtreme Gradient Boosting (XGB) models across four major forest regions in China. Model optimization progressed through three stages: the Unoptimized (UO) XGB model achieved R2 = 0.77 and RMSE = 1.42 g Cm−2 d−1; the Hyperparameter Optimized (HO) XGB model using RGSA improved performance by 5.19% in R2 (0.81) and reduced RMSE by 9.15% (1.29 g Cm−2 d−1); the Hyperparameter and Variable Combination Optimized (HVCO) XGB model with selected variables (LAI, Temp, NR, VPD, and NDVI) further enhanced R2 to 0.83 and decreased RMSE to 1.23 g Cm−2 d−1. The optimized GPP estimates exhibited high spatial consistency with existing high-quality products like GOSIF GPP, GLASS GPP, and FLUXCOM GPP, validating the model’s reliability and effectiveness. This research provides crucial insights for improving GPP estimation accuracy and optimizing ML methodologies for forest ecosystems in China.

Probabilistic analysis of active earth pressures in spatially variable soils using machine learning and confidence intervals

This study introduces a probabilistic framework for assessing active earth pressures in soils exhibiting spatial variability in friction angles and unit weight. These properties are modeled using random fields with log-normal distributions and spatial correlation lengths. Monte Carlo simulations (MCS) are integrated with finite element limit analysis (FELA) to evaluate the failure probability under different design safety factors. To improve computational efficiency and prediction accuracy, machine learning models, such as Multivariate Adaptive Regression Splines (MARS), are utilized to predict failure probabilities based on key spatial variability parameters. A two-phase optimization approach, combining Random Search and Adaptive Sampling, is employed to refine the hyperparameters of the machine learning model. Confidence intervals are incorporated to quantify prediction reliability, providing engineers with robust decision-making tools under uncertainty. Furthermore, adaptive finite element meshes are applied to capture irregular stochastic failure mechanisms, offering deeper insights into the impact of spatial variability. The study produces parametric results in the form of practical contour design charts, aiding engineers in optimizing safety margins while accounting for soil variability. By combining computational methods, machine learning, and uncertainty quantification, this research enhances geotechnical design practices, ensuring more reliable and cost-effective solutions.

Research on Ship Replenishment Path Planning Based on the Modified Whale Optimization Algorithm.

Ship replenishment path planning has always been a critical concern for researchers in the field of security. This study proposes a modified whale optimization algorithm (MWOA) to address single-task ship replenishment path planning problems. To ensure high-quality initial solutions and maintain population diversity, a hybrid approach combining the nearest neighbor search with random search is employed for initial population generation. Additionally, crossover operations and destroy and repair operators are integrated to update the whale's position, significantly enhancing the algorithm's search efficiency and optimization performance. Furthermore, variable neighborhood search is utilized for local optimization to refine the solutions. The proposed MWOA has been tested against several algorithms, including the original whale optimization algorithm, genetic algorithm, ant colony optimization, hybrid particle swarm optimization, and simulated annealing, using traveling salesman problems as benchmarks. Results demonstrate that MWOA outperforms these algorithms in both solution quality and stability. Moreover, when applied to ship replenishment path planning problems of varying scales, MWOA consistently achieves superior performance compared to the other algorithms. The proposed algorithm demonstrates high adaptability in addressing diverse ship replenishment path planning problems, delivering efficient, high-quality, and reliable solutions.

Understanding Electron Beam-Induced Chemical Polymerization Processes of Small Organic Molecules Using Operando Liquid-Phase Transmission Electron Microscopy.

Electron beams evolved as important tools for modern technologies that construct and analyze nanoscale architectures. While electron-matter interactions at atomic and macro scales are well-studied, a knowledge gap persists at the molecular to nano level─the scale most relevant to the latest technologies. Here, we employ operando liquid-phase transmission electron microscopy supported by density functional theory calculations and a mathematical random search algorithm to rationalize and quantify electron beam-induced processes at the molecular level. By examining a series of small organic molecules, we identify critical physical and chemical parameters that dictate polymerization rates under continuous electron beam irradiation. Our findings offer a deeper understanding of electron beam-induced reactions, enabling the prediction of molecular reactivities from a classical chemistry perspective. These insights apply equally to other soft matter systems and, thus, are of fundamental interest to scientists and engineers who use electron beams to analyze or to manipulate nanoscale matter.