A systematic review of hyperparameter optimization techniques in Convolutional Neural Networks

Decision Analytics Journal, Journal Year: 2024, Volume and Issue: 11, P. 100470 - 100470

Published: April 24, 2024

Convolutional Neural Network (CNN) is a prevalent topic in deep learning (DL) research for their architectural advantages. CNN relies heavily on hyperparameter configurations, and manually tuning these hyperparameters can be time-consuming researchers, therefore we need efficient optimization techniques. In this systematic review, explore range of well used algorithms, including metaheuristic, statistical, sequential, numerical approaches, to fine-tune hyperparameters. Our offers an exhaustive categorization (HPO) algorithms investigates the fundamental concepts CNN, explaining role variants. Furthermore, literature review HPO employing above mentioned undertaken. A comparative analysis conducted based strategies, error evaluation accuracy results across various datasets assess efficacy methods. addition addressing current challenges HPO, our illuminates unresolved issues field. By providing insightful evaluations merits demerits objective assist researchers determining suitable method particular problem dataset. highlighting future directions synthesizing diversified knowledge, survey contributes significantly ongoing development optimization.

Language: Английский

---

Flood algorithm (FLA): an efficient inspired meta-heuristic for engineering optimization

The Journal of Supercomputing, Journal Year: 2024, Volume and Issue: 80(15), P. 22913 - 23017

Published: July 1, 2024

Language: Английский

---

A hybrid differential evolution particle swarm optimization algorithm based on dynamic strategies

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Feb. 6, 2025

Particle Swarm Optimization (PSO), a meta-heuristic algorithm inspired by swarm intelligence, is widely applied to various optimization problems due its simplicity, ease of implementation, and fast convergence. However, PSO frequently converges prematurely local optima when addressing single-objective numerical inherent rapid To address this issue, we propose hybrid differential evolution (DE) particle based on dynamic strategies (MDE-DPSO). In our proposed algorithm, first introduce novel inertia weight method along with adaptive acceleration coefficients dynamically adjust the particles' search range. Secondly, velocity update strategy that integrates center nearest perturbation term. Finally, mutation crossover operator DE PSO, selecting appropriate improvement, which generates mutant vector. This vector then combined current particle's best position through crossover, aiding particles in escaping optima. validate efficacy MDE-DPSO, evaluated it CEC2013, CEC2014, CEC2017, CEC2022 benchmark suites, comparing performance against fifteen algorithms. The experimental results indicate demonstrates significant competitiveness.

Language: Английский

---

Research on hybrid strategy Particle Swarm Optimization algorithm and its applications

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Oct. 22, 2024

The increasing complexity and high-dimensional nature of real-world optimization problems necessitate the development advanced algorithms. Traditional Particle Swarm Optimization (PSO) often faces challenges such as local optima entrapment slow convergence, limiting its effectiveness in complex tasks. This paper introduces a novel Hybrid Strategy (HSPSO) algorithm, which integrates adaptive weight adjustment, reverse learning, Cauchy mutation, Hook-Jeeves strategy to enhance both global search capabilities. HSPSO is evaluated using CEC-2005 CEC-2014 benchmark functions, demonstrating superior performance over standard PSO, Dynamic Adaptive Inertia Weight PSO (DAIW-PSO), Hummingbird Flight patterns (HBF-PSO), Butterfly Algorithm (BOA), Ant Colony (ACO), Firefly (FA). Experimental results show that achieves optimal terms best fitness, average stability. Additionally, applied feature selection for UCI Arrhythmia dataset, resulting high-accuracy classification model outperforms traditional methods. These findings establish an effective solution

Language: Английский

---

Enhancing load frequency control and automatic voltage regulation in Interconnected power systems using the Walrus optimization algorithm

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Nov. 13, 2024

This paper introduces the Walrus Optimization Algorithm (WaOA) to address load frequency control and automatic voltage regulation in a two-area interconnected power systems. The are critical for maintaining quality by ensuring stable levels. parameters of fractional order Proportional-Integral-Derivative (FO-PID) controller optimized using WaOA, inspired social foraging behaviors walruses, which inhabit arctic sub-arctic regions. proposed method demonstrates faster convergence improved tie-line stabilization compared recent optimization algorithms such as salp swarm, whale optimization, crayfish secretary bird hippopotamus brown bear teaching learning artificial gorilla troop wild horse optimization. MATLAB simulations show that WaOA-tuned FO-PID improves approximately 25%, exhibits considerable settling time. Bode plot analyses confirm stability with gain margins 5.83 dB 9.61 dB, phase 10.8 degrees 28.6 two areas respectively. system modeling validation showcases superior performance reliability enhancing under step, random step disturbance, nonlinearities like GDC GDB, parameter variations.

Language: Английский

---

An improved artificial rabbits optimization for accurate and efficient infinite impulse response system identification

Decision Analytics Journal, Journal Year: 2023, Volume and Issue: 9, P. 100355 - 100355

Published: Nov. 3, 2023

Identifying models with Infinite Impulse Response (IIR) is crucial in signal processing and system identification. This paper addresses the challenges of IIR model identification by proposing an improved version Artificial Rabbits Optimization (ARO) algorithm called ARO (IARO). The IARO integrates adaptive local search mechanism experience-based perturbed learning strategy as two key enhancements to improve effectiveness ARO. These additions aim address loss accuracy during iterations algorithm's ability exploit promising areas. Four benchmark examples different plants are considered, performance proposed compared existing competitive methods. results consistently demonstrate that outperforms convergence for across all orders systems. Visual analysis, curves, coefficient comparison, statistical metrics comparison validate superiority algorithm. Additionally, Wilcoxon signed-rank test provide further evidence supporting superior IARO. comprehensive analysis showcases efficacy accurately identifying work represents a significant advancement identification, offering methodology accurate efficient modeling.

Language: Английский

---

A Survey of Artificial Hummingbird Algorithm and Its Variants: Statistical Analysis, Performance Evaluation, and Structural Reviewing

Archives of Computational Methods in Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: May 27, 2024

Language: Английский

---

Adapted arithmetic optimization algorithm for multi-level thresholding image segmentation: a case study of chest x-ray images

Multimedia Tools and Applications, Journal Year: 2023, Volume and Issue: 83(14), P. 41051 - 41081

Published: Oct. 11, 2023

Language: Английский

---

Enhancing stochastic optimal power flow with modified cheetah optimizer for integrating renewable energy sources

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: April 24, 2025

In this paper, a modified cheetah optimizer (MCO) algorithm is presented, which has been designed to address the optimal power flow (OPF) problem in grids that utilize renewable energy sources (RES). The issue of uncertainty cost models for wind turbines (WTs) and photovoltaics (PVs), can result overestimation or underestimation RES, addressed by including uncertain value direct these units calculate their accurately. MCO methodology was applied various objective functions such as overall operating cost, voltage deviation, pollutant emissions, loss, were evaluated under different cases. Regarding valve point effect observed case 1, response provided amounts $781.9862. Upon assessing emission costs 2, resultant $810.6655 determined. Considering POZs 3, aggregate $781.7165. minimum network loss recorded 4, 2.0738 MW. By mitigating deviations 5 p.u., incurred exceeds twice preceding case. Furthermore, due its applicability large-scale problems, reserve constraint dynamic economic dispatch chosen an additional test MCO. A backward-forward correction method used correct errors three types reserves, improving solution quality. effectiveness solving practical optimization problems demonstrated results 10-unit 30-unit dispatch, achieving lower values than previously published papers. surpasses 15 top publications at $1,016,361. produced unique $3,048,405.

Language: Английский

---

A crossover-based multi-objective discrete particle swarm optimization model for solving multi-modal routing problems

Decision Analytics Journal, Journal Year: 2023, Volume and Issue: 9, P. 100356 - 100356

Published: Nov. 8, 2023

Passengers must use a set of modes and vehicles to reach their destination in complicated urban structures. Choosing an optimal route is optimization problem for these passengers. This study proposes multi-objective algorithm solve the routing multi-modal network. The network considered this transportation with subway, Bus Rapid Transit (BRT), taxi, walking modes. objective functions determine optimized by considering length, traffic, comfort, safety. We develop Crossover-Based Multi-Objective Discrete Particle Swarm Optimization (CBMODPSO) problem. CBMODPSO has been improved using mutation crossover operators. Artificial Bee Colony (MOABC), Ant (MOACO), Biogeography-Based (MOBBO), Gray Wolf (MOGWO), Non-dominated Sorting Genetic Algorithm-II (NSGA-II) algorithms are used evaluate compare results from algorithm. In addition, compared previous research results. show more repeatable than other algorithms. faster convergence rate able get solution smaller generation number much less time. implemented about one-thirties MOBBO duration. Meanwhile, it reproducibility almost twice MOGWO

Language: Английский

---