Derin Ogrenme Algoritmalarinda Model Testleri: Derin Testler

Deep learning, which is a new area of machine learning, has brought the success of recently developed artificial intelligence applications to very high levels. Most of the algorithms that are proposed in ImageNet contest as well as studies published at computer vision conferences, such as CVPR and ICLR, are based on deep learning. Developing deep learning algorithms depends on high amounts of labeled data and training. Algorithms are architecturally complex and the number of parameters trying to be learned during training is mostly in millions range. Training is performed using iterations and during the iterations a high number of models obtained. These obtained models need to be tested with several different test sets. In this study, the automatic testing of the work models obtained from the results of deep learning algorithm training using multiple datasets, work done for automatic determination of the model to be used as results of these tests and experiences gained will be explained. Anahtar Kelimeler. Deep Learning, Supervised Learning, Test Automation