1 November 2019 in Swarm and Evolutionary Computation
4 December 2018 in Complex & Intelligent Systems
During the last decade, large-scale global optimization has been one of the active research fields. Optimization algorithms are affected by the curse of dimensionality associated with this kind of complex problems. To solve this problem, a new memetic framework for solving large-scale global optimization problems is proposed in this paper. In the proposed framework, success history-based differential evolution with linear population size reduction and semi-parameter adaptation (LSHADE-SPA) is used for global exploration, while a modified version of multiple trajectory search is used for local exploitation. The framework introduced in this paper is further enhanced by the concept of divide and conquer, where the dimensions are randomly divided into groups, and each group is solved separately. The proposed framework is evaluated using IEEE CEC2010 and the IEEE CEC2013 benchmarks designed for large-scale global optimization. The comparison results between our framework and other state-of-the-art algorithms indicate that our proposed framework is competitive in solving large-scale global optimization problems.
4 May 2021 in IEEE Access
During last two decades, Differential Evolution (DE) proved to be one of the most popular and successful evolutionary algorithms for solving global optimization problems over continuous space. Proposing new mutation strategies to improve the optimization performance of (DE) is considered a significant research study. In DE, mutation operation plays a vital role in the performance of the algorithm. Therefore, in this paper, comprehensive analysis of the contributions on basic and novel mutation strategies that were proposed between 1995 and 2020 is presented. A new taxonomy based on the structure of the novel mutations is proposed. Numerical experiments on a set of 30 test problems from the CEC2017 benchmark for 10, 30, 50 and 100 dimensions, including a comparison with classical DE schemes and recent mutations schemes are executed. Furthermore, theoretical, and empirical convergence behavior analysis of all mutations is discussed. The paper also presents many recommendations, guidelines, insights, and suggestions for experienced practitioners and interested researchers in designing and developing effective and efficient DE algorithms to address various optimization problems in different fields.
27 April 2021 in IEEE Access
As optimization algorithms have a great power to solve nonlinear, complex, and hard optimization problems, nature-inspired algorithms have been applied extensively in distinct fields in order to solve real life optimization cases. In this paper, modifications for the recently proposed Gaining-Sharing-Knowledge based algorithm (GSK) are presented for enhancing its performance. Gaining-Sharing-Knowledge algorithm is considered as a perfect example of modern nature-inspired algorithm that considered the human life behavior as a source of inspiration in order to solve optimization problems. GSK algorithm mimics the naturalistic phenomenon of gaining knowledge and sharing knowledge back in human being through two major phasis: junior phase and senior phase. The junior phase is based on generating a set of initial solutions (juniors) to start the search process. Subsequently, all individuals move to the senior phase through the interaction with the surrounding environment and the cooperation with other individuals throughout the search process. The main idea in this presented work is extending and improving the original GSK algorithm through proposing a new idea for adaptively obtain the values for the two significant control parameters in GSK: the knowledge factor and the knowledge ratio. Those two significant parameters have the responsibility of controlling the gaining and sharing knowledge during the junior and senior phases throughout the optimization process. Moreover, an adaptive scheme is presented to control the knowledge rate in order to simulate the gaining and sharing experience throughout the human being life span for a specific population by taking into consideration the diverse nature of any population. a non-linear reduction for the population size is used to improve the GSK algorithm’s performance. Finally, a larger initial value of the initial population size (NP) for all dimensions are applied. The presented algorithm (APGSK) is tested on a very recent benchmark testbed on bound constrained numerical optimization that composed of distinct challenging optimization problems with different dimensions. The presented algorithm (APGSK) showed a superior performance comparing with the original GSK and with other four state-of-the-arts algorithms and the winners of CEC2020 competition.
1 April 2025 in Indonesian Journal of Electrical Engineering and Computer Science
In equestrian sports and veterinary medicine, horse welfare is paramount. Horse tiredness, lameness, colic, and anemia can be identified and classified using deep learning (DL) models. These technologies analyze horse images and videos to help vets and researchers find symptoms and trends that are hard to see. Early detection and better treatment of certain disorders can improve horses’ health. DL models can also improve with new data, improving diagnosis accuracy and efficiency. This study comprehensively evaluates three convolutional neural network (CNN) models to distinguish normal and abnormal horses using the generated horse dataset. For this study, a unique dataset of horse breeds and their normal and abnormal states was collected. The dataset includes mobility patterns from this study’s initial data collection. DL models like CNNs and transfer learning (TL) models (visual geometry group (VGG)16, InceptionV3) were employed for categorization. The InceptionV3 model outperformed CNN and VGG16 with over 97% accuracy. Its depth and multi-level structure allow the InceptionV3 model to recognize characteristics in images of varied scales and complexities, explaining its excellent performance.
6 October 2015 in Computer and Information Science
Location-based services are one of the fastest growing technologies. Millions of users are using these services and sharing their locations using their smart devices. The popularity of using such applications, while enabling others to access user’s location, brings with it many privacy issues. The user has the ability to set his location privacy preferences manually. Many users face difficulties in order to set their preferences in the proper way. One solution is to use machine learning based methods to predict location privacy preferences automatically. These models suffer from degraded performance when there is no sufficient training data. Another solution is to make the decision for the intended user, depending on the collected opinions from similar users. User-User Collaborative Filtering (CF) is an example within this category. In this paper, we will introduce an improved machine learning based predictor. The results show significant improvements in the performance. The accuracy was improved from 75.30% up to 84.82%, while the privacy leak was reduced from 11.75% up to 7.65%. We also introduced an integrated model which combines both machine learning based methods and collaborative filtering based methods in order to get the advantages from both of them.