Unsupervised Representation Learning (URL) is the learning process for the automatic extraction of features or representation from the unlabeled raw data or features are learned without any labeled input data in the system. The significant goal of unsupervised representation learning is to determine the informative, low dimensional feature that seizes some basic structure high dimensional input data without human supervision.

Unsupervised representation learning purely depends on the distribution of the data itself to discover effective information. It also enables a form of semi-supervised learning where the features are learned from the unlabeled data is then utilized for the performance improvement in supervised learning with labeled data set. Several approaches of unsupervised feature learning are k-means clustering, principal component analysis, independent component analysis, local linear embedding, autoencoders, matrix factorization, and unsupervised dictionary learning.

Unsupervised representation learning is capable of handling time series data and spatially distributed data on multiple datasets for the applications tasks such as classification and clustering. In robotics, unsupervised representation learning is applied for autonomous learning and accomplished life-long learning.

Some of the application areas of Unsupervised representation learning are Robotics, healthcare, cyber security, and many more. Recent URL applications are 3D object detection, natural language processing, image detection, recognition and classification, human action detection, speech recognition, defect prediction, navigation, and many more.