Discriminative Autoencoder for Feature Extraction: Application to\n Character Recognition

Conventionally, autoencoders are unsupervised representation learning tools.\nIn this work, we propose a novel discriminative autoencoder. Use of supervised\ndiscriminative learning ensures that the learned representation is robust to\nvariations commonly encountered in image datasets. Using the basic\ndiscriminating autoencoder as a unit, we build a stacked architecture aimed at\nextracting relevant representation from the training data. The efficiency of\nour feature extraction algorithm ensures a high classification accuracy with\neven simple classification schemes like KNN (K-nearest neighbor). We\ndemonstrate the superiority of our model for representation learning by\nconducting experiments on standard datasets for character/image recognition and\nsubsequent comparison with existing supervised deep architectures like class\nsparse stacked autoencoder and discriminative deep belief network.\n