With the availability of a huge amount of data and powerful computational resources, the scope of machine learning algorithms has widened recently. A convolutional neural network (CNN) is one of the most popular machine learning techniques for image classification and pattern recognition tasks.

In this paper, the authors propose an optimized CNN architecture for the age classification problem. They performed an analysis of “the effect of having a different number of convolutional layers and fully connected layers.” Sixteen different CNN architectures having different depth (with a different number of convolutional and fully connected layers) were explored for the task of age classification. The authors used the craniofacial longitudinal morphological face (MORPH) database for conducting the experiments. “Using exact, top-3, and 1-off criterion,” the CNN having four convolutional and two fully connected layers was selected as the optimal architecture for the given task.

The optimal architecture was decided by considering a fixed set of different architectures and testing the performance of each one of them. This approach was feasible, as the number of architectures being evaluated was only 16. However, a much larger number of possible architectures could have been considered by using some metaheuristic algorithm, like a genetic algorithm.

The paper is recommended for students and researchers working in the field of machine learning and exploring applications of CNN for pattern recognition.