Comparing and Analysis of Different Optimization Techniques on Sparse Multi-Class Data

It is a matter of fact that there are certain optimization techniques for sparse data and multi-class data. Optimization techniques generally maximizes or minimizes an error function which depends on the internal parameter of our training model. The effect of different major optimizers on sparse multi-class data can be observed, analyzed and compared based on their loss and accuracy curve. For the analysis, CIFAR-10 data are chosen and trained on a convolutional neural network for different optimizers and then validate on testing data. The objective of this study was to find out the effects of different major optimization techniques in various applications of image processing and state of the art Deep learning. The Result of the study reveals that Nadam will be the best suiting optimization techniques for sparse multi-class datasets.