Classifying and annotating images is an important task in machine learning. Many algorithms have been proposed for these tasks, based on features such as color, texture, and shape. The success of these algorithms is dependent on the selection of features. Deep learning models are widely used to learn abstract, high-level representations from raw data. Energy-based models are the most commonly used deep learning models formed by pre-training the individual restricted Boltzmann machines in a layerwise fashion and then stacking together and training them using error backpropagation. In the deep convolutional neural networks, the convolution operation is used to extract features from different sub-regions of the images to learn better representations. To reduce the time taken for training, models that use convex optimization and kernel trick have been proposed.
We explore some of these deep learning models, for the task of image classification. These models are capable of extracting high-level features from raw images. We examine the quality of the features learned. The performance of these models is compared with that of some of the state-of-the-art models using a set of benchmark datasets. For the task of image annotation, we use a deep convolutional neural network to learn abstract features from raw images and then use them as inputs to the convex deep learning models. The performance of the proposed approach is evaluated on benchmark image datasets.
These are some of the models that were used:
These are some of the results we obtained for the CIFAR dataset.