Objective function in convolutional neural network to generate caption for image - a review

An objective function is one strategy to achieve a fitting model in machine learning. The target of this paper is to acquire more information about the objective function and to observe of it application in CNN. Various CNN architecture has proposed to achieve high accuracy by apply of the objective function. We used the CNN framework as a method to explain the content of architecture. To achieve a good model, every CNN used an objective function as a parameter to measure the closeness between the learning dataset and the actual dataset. As a pre-trained model to extract the critical feature, many scholars proposed a pre-trained CNN model to get high accuracy and a significant model. One of the ablation studies in CNN is a reformulation of the objective function. An objective function has often shown by a matrix operation known as a loss function or loss entropy. As a result, from this research is various CNN architecture models that tailor to many different objects. We can review from architecture, formulation, filter, and dense layer to achieve a good feature extraction as a feature map. Many parameters can observe on every step of CNN. Impact of this review, we can get a baseline model as beginning research to develop a new CNN architecture that can compare with the baseline model.