Time Complexity in Deep Learning Models

The convolution neural network is gaining a lot of popularity in image classification problems nowadays. It has been used in many different classification problems, like medical imaging, handwritten digits, image classification, etc. It is very critical to estimate the time required by the model to achieve the desired task. Earlier studies have used CNN for different perspectives and given the depth of each layer of CNN, but estimating the time and space taken by these layers is missing. This paper's prime idea is to find a CNN model's time complexity. The present work involves computational studies to find the factors that affect the model's performance, the time each layer takes to run, and how it affects the model's overall performance. Time complexity has been discovered on eight different models, varying by the size of filters, number of convolutional layers, number of filters, number of fully connected layers, and kernel size. The result shows that factors like an optimizer, batch size, filter, and neurons greatly impact the time taken by the model. The convolutional layers, max pool, and fully connected layers directly affect the performance of the model.