Increasing the object recognition performance of deep learning algorithms with hyperparameter optimization

The rapid development of technology is also reflected in the field of deep learning. There have been improvements in the number and performance of Deep Learning algorithms. Along with this, there has been an increase in the number of hyperparameters used in algorithms. Hyperparameters have an important place during the training of algorithms. Ideal determination of hyperparameters affects the training process and performance rates. This means cost and time in the training process. Hyperparameter values are determined before deep learning algorithms start to run. Determining hyperparameters is a very laborious process. In this study, SGD, Adam, NAdam and RMSProp hyperparameter optimization techniques will be used when determining the hyperparameters in deep learning algorithms. These hyperparameter optimization techniques will be applied on the data set in 50, 100, 200 and 300 iteration times. As a result of the application, it is aimed to automatically determine the hyperparameters and keep them at an ideal level, reduce the duration of the training and increase the performance. The hyperparameter and performance values obtained as a result of the study will be compared and the most ideal hyperparameter values will be determined. The brain tumor data set obtained from the Roboflow site will be used to verify the performance of the hyperparameter optimization techniques specified on the Yolov8 algorithm, which is among the deep learning algorithms.