Convolutional Neural Network Architecture Design by the Tree Growth Algorithm Framework

摘要: This paper presents tree growth algorithm framework for designing convolutional neural network architecture. Convolutional networks are a special class of deep that typically consist several convolution, pooling and fully connected layers. have proved to be robust method tackling various image classification tasks. One the most important challenges from this domain is find architecture has best performance specific application. The depends on set hyper-parameter values such as number dense layers, kernels per layer kernel size. Optimization hyperparameters was performed by novel belongs group swarm intelligence metaheuristics. robustness, solutions quality proposed validated against well-known MNIST dataset. Conducted comparative analysis demonstrated frameworks obtains promising results in domain.