Analysis of Performance for a Chairs Classifier through Deep Learning Javier Maldonado Romo, Mauricio Olguín-Carbajal, Israel Rivera-Zárate, Raul Galvan Instituto Politécnico Nacional, Mexico City, Mexico javier.mr.21@gmail.com, molguinc@ipn.mx, irivera@ipn.mx, raulgalvan92@outlook.com Abstract. Deep learning is a branch of machine learning and this technique allows us to create classifiers. We must find the best dataset size for a classifier process to permit using less time and give good accuracy. In this paper we will propose models with different deep layers and size dimensions for detecting the best model to solve a task that needs quick time processing. Keywords: Artificial intelligence, deep learning, convolutional neural network, classifier. 1 Introduction In this topic of investigation about artificial intelligence there are many techniques for processing and classifying the information. Each technique performs in different situations. Recently, there has been an interest in topics regarding artificial intelligence that is deep learning. Deep learning is a topic that is not new but has one feature that allows it to be a good technique to detect patrons in photos, audios and linguistics. Deep learning is a popular technique, but it still needs graphic card units to improve performance. GPU helps to improve processing times but it is still necessary to use deep learning because the GPU has many computer process units that work in parallel to solve the problem faster. There are web sites which contains several datasets concerning different subjects. One dataset that is famous from deep learning is the CIFAR-10 dataset. This dataset is found on the Kaggle web site. It is a competition where there are ten classes that contain one thousand images by class; the images are in red, blue and green channel colors. Also it has the dimensions of thirty two in width and height. In 2009 one research team achieved an accuracy of 92% using deep learning. Many research teams around the world use deep learning for processing data using GPU. The time is less compared with CPU and cheaper than CPU cluster. To apply deep learning we can use the following tips: the first aim is that dataset has much information, the different classes have many images that represents a split and the system could process the information because dataset has much data that 149 ISSN 1870-4069 Research in Computing Science 118 (2016) pp. 149–156; rec. 2016-09-27; acc. 2016-10-28