Designing Convolution Neural Network Architecture by utilizing the Complexity Model of the Dataset

Convolutional Neural networks (CNN) have been utilized in a wide variety of areas, with a high degree of performance. The design of the CNN has been a general problem addressed mainly by using the experts to design the architecture of the CNN model, which is quite inefficient. In recent times there has been the need to develop the CNN architecture automatically, resulting in multiple approaches that have been utilized for this process. In this paper, we give an overview of the multiple approaches utilized for automatic CNN architecture design and also propose a method to design the architecture, which utilizes the dataset itself for the process of designing the architecture of the CNN. The method shows good performance and in terms of the resources required to generate the architecture has significantly better performance than state-of-the-art models.