A Wide Analysis of Loss Functions for Image Classification Using Convolution Neural Network
Janaki Raman Palaniappan
Abstract
Thanks to the humongous availability of data. The huge amount of data is helping the Deep Learning, continuously grow rapid in all kinds of domains and industries. Convolution Neural Network (CNN) is a class of Artificial Neural Network (ANN) which is a method of Deep Learning that has delighted across many fields that includes computer vision domain. The Deep Learning mimics the human brain as it memorizes the information and detect by patterns. Classifying the images is done by training the CNN model. As a human we make mistakes sometimes, not identifying the image due to similar features of a different object. Even CNN model makes mistakes which can be termed as ‘loss of detection’. A loss functions generally compares expected and predicted values during the training phase. Here the aim is to minimize the loss for a neural network model to have a better accuracy as it turns the model better resultant. In this paper, different kinds of loss functions have been analyzed deeply to understand which loss function performs better and/or improves accuracy for a multiclass classification particularly for images.