Abstract:Numerous optimization problems can be addressed using metaheuristics instead of deterministic and heuristic approaches. This study proposes a novel population-based metaheuristic algorithm called the Exponential Distribution Optimizer (EDO). The main inspiration for EDO comes from mathematics based on the exponential probability distribution model. At the outset, we initialize a population of random solutions representing multiple exponential distribution models. The positions in each solution represent the exponential random variables. The proposed algorithm includes two methodologies for exploitation and exploration strategies. For the exploitation stage, the algorithm utilizes three main concepts, memoryless property, guiding solution and the exponential variance among the exponential random variables to update the current solutions. To simulate the memoryless property, we assume that the original population contains only the winners that obtain good fitness. We construct another matrix known as memoryless to retain the newly generated solutions regardless of their fitness compared to their corresponding winners in the original population. As a result, the memoryless matrix stores two types of solutions: winners and losers. According to the memoryless property, we disregard and do not memorize the previous history of these solutions because past failures are independent and have no influence on the future. The losers can thus contribute to updating the new solutions next time. We select two solutions from the original population derived from the exponential distributions to update the new solution throughout the exploration phase. Furthermore, EDO is tested against classical test functions in addition to the Congress on Evolutionary Computation (CEC) 2014, CEC 2017, CEC 2020 and CEC 2022 benchmarks, as well as six engineering design problems. EDO is compared with the winners of CEC 2014, CEC 2017 and CEC 2020, which are L-SHADE, LSHADE−cnEpSin and AGSK, respectively. EDO reveals exciting results and can be a robust tool for CEC competitions. Statistical analysis demonstrates the superiority of the proposed EDO at a 95% confidence interval.

Extra dimension algorithm: a breakthrough for optimization and enhancing DNN efficiency

Massive Dimensions Reduction and Hybridization with Meta-heuristics in Deep Learning

EDD: Efficient Differentiable DNN Architecture and Implementation Co-search for Embedded AI Solutions

An Enhanced Dynamic Differential Annealed Algorithm for Global Optimization and Feature Selection

A methodological framework for optimizing the energy consumption of deep neural networks: a case study of a cyber threat detector

An algorithmic framework for the optimization of deep neural networks architectures and hyperparameters

Exponential distribution optimizer (EDO): a novel math-inspired algorithm for global optimization and engineering problems

An Efficient Optimization Technique for Training Deep Neural Networks

Towards Hyperparameter-Agnostic DNN Training via Dynamical System Insights

Energy-efficient offloading for DNN-based applications in edge-cloud computing: A hybrid chaotic evolutionary approach

EEE, Remediating the failure of machine learning models via a network-based optimization patch

EINNET: Optimizing Tensor Programs with Derivation-Based Transformations.

U-AEFA: Online and offline learning-based unified artificial electric field algorithm for real parameter optimization

A Convergent ADMM Framework for Efficient Neural Network Training

Towards Ultra-High Performance and Energy Efficiency of Deep Learning Systems: An Algorithm-Hardware Co-Optimization Framework

NGDE: A Niching-Based Gradient-Directed Evolution Algorithm for Nonconvex Optimization

An Equivalent Circuit Approach to Distributed Optimization

Daydream: Accurately Estimating the Efficacy of Optimizations for DNN Training

AA-DLADMM: An Accelerated ADMM-based Framework for Training Deep Neural Networks

Enhancing Deep Learning with Optimized Gradient Descent: Bridging Numerical Methods and Neural Network Training

A dynamic stochastic search algorithm for high-dimensional optimization problems and its application to feature selection