Attention-Dependent and Continuous Representation Learning using Deep Autoencoders