Representation learning is the learning representation of input data sets to perform the task better by transforming it or extracting features from it. Representation learning approaches allow the system to determine the representation for prediction or classification tasks from raw data. Representation learning helps in understanding the overall behavior of the model by reducing the dimensions or discovering the patterns.

Representation learning methods are supervised learning, unsupervised learning, and deep architectures. Supervised representation learning approaches are Supervised dictionary learning and neural networks. Unsupervised representation learning approaches are K-means clustering, principal component analysis, local linear embedding, independent component analysis, and unsupervised dictionary learning.

Deep architectures are widely used in representation learning due to their insensitivity to complex noise or data conflicts. Deep neural network approaches in representation learning are Multilayer perceptron, Convolutional neural network, Restricted Boltzmann machine, and Autoencoder. Problems overcome by deep representation learning are models with less training data and more computation complexity. Applications of deep representation learning are speech recognition, image analysis, natural language processing, and game playing.